New Your team’s decisions, in one playbook every coding agent works from. Never answer your agent twice
// SCAN REPORT

Visionmcp

376 tools. 190 can modify or destroy data without limits.

2 destructive tools with no built-in limits. Policy required.

Last updated:

190 can modify or destroy data
186 read-only
376 tools total

Community server · catalogue entry checked 30/07/2026

How to control Visionmcp ↓

TOOL MAP

What Visionmcp exposes to your agents

Read (186) Write / Execute (188) Destructive / Financial (2)

Running more than one server? Check your whole stack →

MOST DANGEROUS TOOLS
Critical Risk

The most dangerous Visionmcp tools

Destructive · High active_learning.rollback Rolling back a revision replaces the current state with a predecessor, effectively destroying the current revision's changes. Destructive · Medium artifact.reap_stale_uploads The tool permanently expires/deletes abandoned partial uploads. Execute · High active_learning.execute_retraining This tool initiates a training pipeline execution. Execute · High active_learning.plan_retraining This tool triggers an external operation (model retraining) whose effects are significant and depend on arguments (the correction dataset, benchmark, and training parameters).

190 of Visionmcp's 376 tools can modify, destroy, or commit something on every call — and an agent calls them with no built-in limits.

POLICY

How to control Visionmcp

PolicyLayer is an MCP gateway — it sits between your AI agents and Visionmcp, and nothing reaches the server without passing your rules. These are the rules we recommend:

Deny destructive operations
{
  "active_learning.rollback": {
    "deny_if": [
      {
        "conditions": [],
        "on_deny": "Blocked by default. Requires approval."
      }
    ]
  }
}

Destructive tools should never be available to autonomous agents without human approval.

Rate limit write operations
{
  "active_learning.promote": {
    "limits": [
      {
        "counter": "active_learning.promote_per_hour",
        "window": "hour",
        "max": 30,
        "scope": "grant"
      }
    ]
  }
}

Prevents bulk unintended modifications from agents caught in loops.

Cap read operations
{
  "active_learning.compare": {
    "limits": [
      {
        "counter": "active_learning.compare_per_minute",
        "window": "minute",
        "max": 60,
        "scope": "grant"
      }
    ]
  }
}

Controls API costs and prevents retry loops from exhausting upstream rate limits.

  1. Create a free account and register Visionmcp — nothing to install.
  2. Add these rules — paste them, or build them visually. Tune the limits to your setup.
  3. Point your MCP client (Claude, Cursor, anything) at your gateway URL.
ENFORCE POLICY ON VISIONMCP →

Instant setup, no code required.

FULL CATALOGUE

All 376 Visionmcp tools

DESTRUCTIVE 2 tools
Destructive active_learning.rollback Receipt-roll back the current revision and restore its predecessor. Destructive artifact.reap_stale_uploads Expire abandoned partial uploads while preserving registered artifacts.
EXECUTE 68 tools
Execute active_learning.execute_retraining Queue the exact training run created by an artifact-bound active-learning plan. Execute active_learning.plan_retraining Create a correction dataset and offline training run against a fixed benchmark. Execute benchmark.bootstrap_calibration Generate, review, and require a verified accepted L3 synthetic calibration project. Execute benchmark.bootstrap_dgx_spark Bootstrap an unaccepted, measured DGX Spark reconstruction candidate. Execute benchmark.bootstrap_rtx_5090_fe Bootstrap an unaccepted, measured RTX 5090 FE reconstruction candidate. Execute benchmark.refine_dgx_spark_base_foot_candidate Queue a recessed DGX foot refinement without changing the 150 mm body envelope. Execute benchmark.refine_dgx_spark_visual_candidate Queue a material and rear-I/O DGX candidate while retaining exact body dimensions. Execute benchmark.refine_rtx_5090_fe_front_frame_candidate Queue an evidence-bound RTX front X-frame refinement at strict dimensions. Execute benchmark.refine_rtx_5090_fe_visual_candidate Queue a seven-blade RTX visual candidate while retaining strict dimensions. Execute blender.generate_lod Queue a non-destructive, audited Blender LOD checkpoint for named mesh objects. Execute blender.prepare_asset Queue bounded retopo, UV, PBR, bake, rig, animation, LOD, collision, and repair. Execute blender.render Queue safe headless renders for the current evidence-bound camera solution. Execute camera.derive_undistorted Derive source-linked pinhole frames while preserving pose and scale authority. Execute campaign.advance Advance exactly one validated OBSERVE-to-ROLLBACK controller state. Execute campaign.resume Resume a paused campaign at its preserved controller state. Execute campaign.start Start a persistent evidence-governed reconstruction controller. Execute candidate.evaluate_transaction Evaluate every mandatory gate atomically and auto-reject any regression. Execute component.generate Queue allowlisted Blender generation into a new audited checkpoint. Execute coverage.acquire_missing Execute a bounded governed search for gaps, or issue precise capture requests. Execute dataset.generate Queue safe headless Blender rendering for a planned synthetic dataset. Execute dataset.train_feature_model Plan and queue offline feature-model training; no weights are downloaded. Execute evidence.discover Execute reviewed search queries and register results with unresolved source rights. Execute geometry.run_ensemble Queue independent geometry hypotheses followed by license-aware consensus. Execute photogrammetry.run Queue an evidence-bound photogrammetry lane without promoting unresolved scale. Execute portfolio.execute_initial Execute cheap local portfolio lanes and record unavailable workers truthfully. Execute portfolio.execute_parametric_seed Build a fresh editable dimension-bound seed with no private starting model. Execute recon.bootstrap Queue the evidence-through-render bootstrap without claiming final fidelity. Execute recon.continue_multiview_search Queue or resume isolated render-and-compare evaluation for a planned search. Execute reference.propose_masks Generate replayable automatic mask proposals with no review authority. Execute repair.apply Queue an approved repair, checkpoint audit, topology/ray validation, and rear render. Execute vision.acquire.quarantine Advance verified bytes through license, rights, and static scan. Execute vision.app.structure Build approximation-only structure, UI-state, and behaviour graphs. Execute vision.benchmark_target Run the owned application benchmark; external gates report blocked rather than pass. Execute vision.binary.analyze Run one bounded headless Ghidra analysis in an isolated worker process. Execute vision.build_industrial_model Build an editable unbranded industrial-design hierarchy with geometric residual authority. Execute vision.build_reconstruction_portfolio Run requested reconstruction backends and seal a ReconstructionPortfolio. Execute vision.compile Compile observed layout/text/paint evidence into deterministic HTML/CSS. Execute vision.compile_cinematic_scene Export browser motion table from a CameraPathGraph; optional Blender bake is honest. Execute vision.compile_motion Compile MotionGraph into reusable sticky, scroll, reveal, media, camera, and responsive primitives. Execute vision.evaluate Run a capsule evaluation or mandatory global frontend non-regression gate. Execute vision.fit_parametric_model Fit a RANSAC plane/box/cylinder/sphere; result authority is MODEL_DERIVED. Execute vision.generate_fur Groom fur in Blender; returns measured groom report or an explicit blocked state. Execute vision.generate_lods Generate measured LODs; report blender_used=false when Blender is blocked. Execute vision.generate_texture_set Generate tileable PBR maps from a MaterialHypothesis (not observed textures). Execute vision.generate_uvs Generate UVs in Blender; return measured quality or a blocked state. Execute vision.generate.authorized Generate after verified search exhaustion or an explicit original declaration. Execute vision.measure_visual_parity Run governed pixel, perceptual, layout, typography, crop, stacking, and distribution metrics. Execute vision.retopologize Retopologize via real Blender; blocked state is reported if Blender cannot run. Execute vision.run Queue normalized geometry evidence generation without promoting unresolved scale. Execute vision.run_perceptual_critics Run specialist critics and seal a PerceptualCritique with measured findings. Execute vision.saccade Plan and commit a saccade to a new region (IOR-aware). Execute vision.solve_calibration_board Recover metric OpenCV cameras from an authoritative measured chessboard. Execute vision.solve_cameras Queue camera solving while preserving backend authority and uncertainty labels. Execute vision.solve_lighting Solve a LightingHypothesisSet capped by derive() over inputs. Execute vision.solve_product_camera Solve an evidence-bound product camera from six or more 3D/2D landmarks. Execute vision.track Associate detections to tracks for one frame (IoU + appearance + Kalman). Execute vision.transplant_feature Compile and verify a clean-room behavior capsule without copying source assets. Execute visual_oracle.train Plan and queue camera-bound offline appearance-oracle training. Execute worker.fail Report a structured failure and retry up to the job's configured attempt limit. Execute workflow.audit_reference_fidelity Run the first complete audit slice asynchronously from evidence through receipt. Execute workflow.continue_autonomous Execute the next safe evidence-derived action or stop at an authority boundary. Execute workflow.fit_component Queue a robust fit proposal; named review is required before parameters change. Execute workflow.generate_original_asset Launch an explicitly L0 generated-design portfolio without measured-target claims. Execute workflow.reconstruct_hardware Launch the computer-hardware ontology and reconstruction portfolio. Execute workflow.reconstruct_organic_subject Launch anatomical landmarks and distinguish real reconstruction from generation. Execute workflow.reconstruct_packaging Launch packaging folds, layers, print regions, and reconstruction hypotheses. Execute workflow.reconstruct_product Queue the complete staged reconstruction/audit workflow and retain all blockers. Execute workflow.reconstruct_vehicle Launch the vehicle ontology, constraints, search plan, and candidate portfolio.
WRITE 120 tools
Write active_learning.promote Activate a commercially eligible non-regressing checkpoint after named review. Write active_learning.record_corrections Bind named human corrections to requested predictions. Write artifact.abort_upload Abort the authenticated worker's incomplete artifact transfer. Write artifact.begin_upload Begin a sequential digest-declared worker output upload. Write artifact.complete_upload Verify declared size/SHA-256 and atomically register a worker output artifact. Write artifact.upload_chunk Append one authenticated base64 chunk to a declared output transfer. Write benchmark.beast_audit Audit a Beast Mode stage from persisted evidence and record incomplete gates. Write benchmark.review_dgx_foam_lod Record a named, receipt-backed DGX foam LOD policy approval. Write benchmark.revise_rtx_5090_fe_candidate Queue a distinct strict-v2 RTX candidate with a measured 137 × 40 × 304 mm envelope. Write blender.export_blend Queue a portable editable Blender export of the authoritative accepted scene. Write campaign.pause Pause a campaign while preserving its exact state and evidence. Write campaign.progress Persist host-visible progress without advancing the controller state. Write candidate.accept Accept only a candidate backed by its own passed transaction receipt. Write candidate.reject Reject a recoverable candidate with a named transition receipt. Write category.select Select the strongest category pack for the canonical target. Write component.create Create a typed parametric component bound to project evidence. Write component.repair Create a governed component-local repair proposal without mutating the baseline. Write component.review_fit Apply or reject a proposed fit through a named, revision-checked decision. Write component.update Create a new component revision by updating named parameters. Write context.create_packet Create a compact component-local decision packet instead of sending the project. Write coverage.observe_surface Bind one governed observation to a canonical surface-atlas cell. Write dataset.import_training_result Import a hash- and license-bound checkpoint produced by an approved worker. Write evidence.acquire Register provenance and rights, then ingest a permitted local working reference. Write evidence.propose_legacy_reference_adoption Propose artifact-bound migration of legacy references without inferring rights. Write evidence.register_search_provider Register a bounded search API after named terms, privacy, and secret review. Write evidence.review_conflict Exclude, branch, or explicitly resolve one classified evidence conflict. Write evidence.review_legacy_reference_adoption Adopt or exclude one legacy reference through a named immutable review. Write feature.add Add an evidence-bound technical feature to the project graph. Write feature.link Link a cross-view observation to a technical feature with named provenance. Write feature.review Record an explicit named approval or rejection for one technical feature. Write generative3d.generate Create and optionally queue a licensed artifact-bound generative proposal. Write generative3d.import_result Import hash-bound generative outputs as non-acceptance-eligible hypotheses. Write material.create Propose an evidence-bound appearance profile with no geometry authority. Write material.review Record a named material-profile approval or rejection. Write measurement.add Record a typed measurement with evidence class, uncertainty, and references. Write measurement.add_physical Register calibrated physical inputs while retaining uncertainty and calibration. Write measurement.bind_source_provenance Bind a manufacturer measurement to a governed source without reviewing its value. Write measurement.calibrate Record a reviewed x/y/z calibration envelope as authoritative measured evidence. Write measurement.correct Correct metric evidence while retaining the prior value and named audit rationale. Write measurement.grid_create Create a normalized perspective grid with rulers, targets, snapping, and links. Write measurement.link Link a stored measurement to reference evidence with named audit provenance. Write model.approve_source Approve a model URL, license, and expected digest without downloading it. Write model.import_checkpoint Import a manually acquired checkpoint only when its approved digest matches. Write portfolio.fusion_plan Plan evidence-safe candidate fusion without treating generated topology as measured. Write portfolio.generate Create bounded classical, learned, generative, and semantic hypotheses. Write portfolio.record_result Attach evidence outputs and scores to one isolated portfolio candidate. Write project.create Create a portable project (millimetres, right-handed, Z-up). Write receipt.export Queue a machine-verifiable acceptance receipt without overstating fidelity. Write recon.repair Propose a supported repair; a separate named approval is required before application. Write recon.review_optimization Apply or reject an optimization proposal through named revision-checked review. Write reference.import Queue immutable reference ingestion with metadata and quality inspection. Write reference.import_reviewed_mask Bind a named, reviewed binary silhouette mask to an immutable image reference. Write reference.review_mask_proposal Accept or reject an automatic mask through a named immutable decision. Write repair.approve Authorize safe checkpoint evaluation; this does not accept generated geometry. Write repair.propose_mac_studio_grille Create an evidence-bound Mac Studio rear-grille proposal without editing geometry. Write repair.review Record final repair acceptance or rejection; acceptance requires a clean receipt. Write review.model_tier Record a tier decision; acceptance cannot exceed a valid receipt's fidelity. Write review.request_capture Record a named, coverage-driven request for an additional capture view. Write role.assign Assign one persistent advisory role task by uncertainty, cost, and objective. Write role.complete Complete a role handoff as advisory evidence without accepting or promoting scenes. Write role.waiting Mark an advisory role task as waiting for explicit external input. Write semantic_model.bind Bind Blender objects and references to one stable semantic component ID. Write semantic_model.bootstrap Create a semantic digital-twin graph whose operations use stable semantic IDs. Write semantic_model.extend Idempotently add target-specific component types to an existing semantic twin. Write semantic_model.review Accept, reject, or declare a semantic component inapplicable with named review. Write synthetic_view.register Register a consistency-scored hypothetical view that cannot establish acceptance. Write system.warm_service_update Record optional warm-service state for resource-aware reuse and eviction. Write target.resolve Resolve and persist one canonical variant without merging incompatible evidence. Write validation.register_visual_oracle Register a Gaussian/neural appearance oracle with hashed camera/training config. Write validation.supersede_duplicate_comparison Receipt-supersede a replay-identical duplicate comparison without deleting history. Write video.ingest Ingest video, extract frames, classify motion, and select evidence keyframes. Write vision.acquire.import Import parser-passed quarantined content without executing it. Write vision.acquire.resume Resume an incomplete bounded download with Range verification. Write vision.analysis.authorize Create an explicit, optionally target-bound authorization record. Write vision.binary.ingest Hash and register one authorized regular file. Nothing is executed. Write vision.close_project Forget a project path from the session map (does not delete files). Write vision.compile_web_scene Seal a DeliveryManifest with measured budgets; violations are recorded, never raised. Write vision.consolidate_camera_solutions Consolidate disjoint hypotheses into one immutable, still-unapproved camera set. Write vision.create_benchmark Package a licensed source-known benchmark with explicit source isolation. Write vision.generate_replacement_media Generate original role-preserving media from statistics without accepting target pixels. Write vision.import_camera_solution Import a validated data-only manual or calibrated camera solution as unapproved. Write vision.import_depth Ingest a depth map and seal an ObservationBundle at the declared authority. Write vision.import_feature_detections Import model detections as unapproved evidence-bound technical features. Write vision.import_geometry_evidence Import artifact-bound output from an approved external geometry worker. Write vision.import_point_cloud Load a PLY point cloud and seal an ObservationBundle preserving source authority. Write vision.index.add Add one confined file to the classical multimodal index. Write vision.promote_candidate Promote a V2 record via V2Record.promote(); refuses system/auto reviewers. Write vision.propose_pnp_landmarks Persist machine-proposed image/model landmarks without granting authority. Write vision.propose_pnp_landmarks_from_renders Propose hash-bound render/image landmarks or persist an explicit refusal. Write vision.reconstruct Materialize an editable candidate without promoting it to accepted authority. Write vision.refine_camera Queue bounded silhouette camera refinement as an unapproved non-metric proposal. Write vision.repair Create a source-mapped repair queue from independently measured residuals. Write vision.review_camera_solution Explicitly approve a camera solution after its quality evidence has been reviewed. Write vision.review_pnp_landmarks Create an immutable named decision for every proposed correspondence. Write vision.search.archives Register historical page/release metadata under archive terms review. Write vision.search.assets Register asset-family candidates and retain source rights. Write vision.search.code Register code-search leads with GitHub source and license lineage. Write vision.search.exhaustion Seal the mandatory SearchExhaustionReceipt used by generation gates. Write vision.search.images Register image-search metadata without accepting remote image rights. Write vision.search.network_assets Register bounded browser-network asset metadata from an official page. Write vision.search.official Register first-party public source leads without auto-promoting rights. Write vision.search.page_resources Register public page-resource metadata with private-reference default rights. Write vision.search.plan Create and seal the mandatory tier-0-through-tier-7 search plan. Write vision.search.video Register video-search metadata without fetching media bodies. Write vision.update_world_model Incorporate one frame observation into an existing world (append-only). Write visual_geometry.bind_scene Create object-level provisional semantic bindings and PART_OF assembly edges. Write visual_geometry.create_component_packet Create native-reference and diagnostic-render crops for one bound component. Write visual_geometry.create_rig Freeze a receipt-bound review rig; unapproved cameras remain diagnostic. Write visual_geometry.evaluate Create a replayable projection, edge, local, and perceptual scorecard. Write visual_geometry.freeze_baseline Receipt the immutable scene, rig, render, scorecard, export, and benchmark baseline. Write visual_geometry.relate_components Add a governed provisional mechanical relationship between bound objects. Write visual_geometry.repair_degenerate_candidate Create a guarded topology-repair candidate without accepting or promoting it. Write visual_geometry.repropose_binding Supersede only an unreviewed machine binding after a classifier correction. Write visual_geometry.review_binding Named review or acceptance of one receipt-bound visible-object binding. Write worker.complete Complete a leased job only after every declared output artifact is registered. Write worker.register Enroll a capability-advertising worker and return its token exactly once. Write workflow.deliver_promoted Queue verified BLEND, GLB, and receipt delivery for a safely promoted scene. Write workflow.reconstruct_from_public_evidence Create a project and launch the autonomous public-evidence twin workflow. Write workflow.reconstruct_from_user_capture Create a project from owner-supplied evidence and launch reference reconstruction. Write workflow.repair_existing_model Create an evidence-bound repair proposal; authorization and review remain separate.
READ 186 tools
Read active_learning.compare Recompute non-regression from two stored fixed-benchmark evaluations. Read active_learning.start Rank low-confidence, high-impact predictions for governed correction. Read active_learning.start_from_workspace Rank evidence-bound workspace uncertainty for named correction. Read appearance.compare Report governed appearance and full-object comparison gates. Read artifact.describe Describe an input artifact after authenticating the requesting worker. Read artifact.read_chunk Read at most 1 MiB from an immutable input artifact for remote execution. Read benchmark.bootstrap_external_perseverance Bootstrap Stage 4 from reviewed NASA sources and no private starting model. Read blender.export Queue safe headless GLB export of the authoritative imported Blender scene. Read blender.inspect Queue safe headless inventory of the latest imported Blender scene. Read camera.freeze Verify that every camera in a stored solution is a complete immutable snapshot. Read campaign.status Return controller state, iteration, budget, events, and stopping evidence. Read campaign.stop Stop a campaign explicitly and preserve its terminal reason. Read candidate.evaluate_fixed_camera_regression Verify paired fixed-view artifacts and reject an aggregate silhouette regression. Read candidate.promote Promote an accepted scene using its passed evaluation and supersede the baseline. Read candidate.rank Rank stored candidate transactions with passed non-regressing results first. Read category.list List built-in domain ontologies, constraints, and parametric constructs. Read component.fit Queue a robust evidence-bound scalar parameter fit without applying it. Read component.reconstruct Generate semantic parametric components into an isolated audited candidate. Read coverage.analyze Analyze acquired directional evidence and identify missing surfaces. Read coverage.observe_governed_source Credit inspected surface regions to one acquired, named-reviewed source. Read dataset.evaluate Evaluate stored feature predictions with precision, recall, F1, and mask IoU. Read dataset.plan_synthetic Plan a deterministic synthetic technical-product dataset with perfect-label outputs. Read evidence.acquire_url Acquire one pre-reviewed URL with robots, network, size, and provenance guardrails. Read evidence.audit Audit source provenance and rights-ledger completeness. Read evidence.deduplicate Audit exact, perceptual, and mirrored duplicates without deleting evidence. Read evidence.list_legacy_reference_adoptions List legacy-reference adoption proposals and their review state. Read evidence.resolve_conflicts Detect source variants that conflict with the canonical project identity. Read evidence.review_governance Complete a named source-terms, privacy, and rights review. Read evidence.search Generate a category-aware, rights-governed acquisition query plan. Read geometry.compare Report dimensional, semantic, topology, and candidate-transaction geometry gates. Read geometry.validate_glb Validate one project-confined GLB without fetching external resources. Read job.cancel Request cooperative cancellation of a queued or running job. Read job.status Get structured job state, result, error, and event history. Read material.compare Report material-region evidence, calibration, confidence, and review gates. Read material.list List material properties, appearance evidence, confidence, and review state. Read measurement.grid_list List perspective-grid records and their uncertainty and scale evidence. Read model.list List model approvals/installations and the no-silent-download policy. Read portfolio.rank Rank evaluated hypotheses while preferring an editable semantic target. Read project.audit Queue an authoritative scene audit with canonical-unit and safe-mode findings. Read project.open Open and migrate a portable project, returning its canonical resolved path. Read project.status Return project status and directory health. Read recon.audit_existing Queue an authoritative audit of the current imported or generated scene. Read recon.list_multiview_searches List persistent multiview searches, attempts, results, and receipt state. Read recon.next_best_view Return current uncertainty coverage and actionable missing-view guidance. Read recon.optimize Create a multi-objective optimization proposal without mutating the component. Read recon.optimize_multiview Propose a fixed-camera component fit from artifact-bound multiview residuals. Read recon.solve Queue independent camera and normalized geometry hypotheses for later consensus. Read recon.start_multiview_search Plan an idempotent bounded multiview search without mutating the component. Read reference.audit_derivations Audit immutable derived-reference lineage and inherited source governance. Read reference.coverage Create a coverage and next-best-view report for current image references. Read reference.extract_pdf Preserve a PDF and import bounded, source-linked page images through Poppler. Read reference.extract_video Preserve a source video, extract ordered image evidence, and rank frame candidates. Read reference.list_mask_proposals List automatic mask proposals and their review state. Read reference.list_masks List approved, hash-bound reference masks and their named review provenance. Read reference.select_frames Rank decodable image references by deterministic quality signals for human review. Read review.snapshot Return the complete read-only review model used by the local browser application. Read role.list List persistent role-task handoffs in priority order. Read system.capabilities List backend states, licenses, hardware, and outputs without downloading weights. Read system.doctor Core doctor: offline, no optional dependency required. Read system.resource_profiles Discover hardware and publish Compact through Distributed Beast profiles. Read system.warm_service_evict Evict lowest-reuse warm services until the requested memory is available. Read system.warm_service_list List persistent optional worker state without starting any process. Read uncertainty.audit Distinguish observed, measured, inferred, unseen, and synthetic project claims. Read validation.acceptance Evaluate every current fidelity gate and emit a verifiable acceptance receipt. Read validation.compare Queue render, silhouette residual, and metric comparison for imported references. Read validation.coverage Queue a coverage and next-best-view report. Read validation.plan_locality Plan only affected views, passes, regions, and metrics for a semantic change. Read validation.recompute_legacy_comparison Recompute immutable inputs and supersede unreproducible legacy comparison metrics. Read video.extract_keyframes Reanalyze extracted frames into governed keyframes, motion classes, and tracks. Read vision.acquire.download Download to read-only quarantine with redirects, limits, and lineage. Read vision.acquire.extract Safely extract bounded ZIP/TAR archives without links or traversal. Read vision.acquire.preview Validate HTTPS, host allowlist, and public DNS without downloading. Read vision.acquire.verify Verify digest, size, no-execute state, MIME, and magic. Read vision.adapters List installed sensor adapters without launching a browser. Read vision.analysis.toolchain Probe version-only surfaces or return an inert, sandbox-required argv plan. Read vision.analyze_motion Return observed motion tracks, compiled curves, and replay bounds. Read vision.analyze_typography Measure rendered font roles, fallback/load state, metrics, baselines, and line breaks. Read vision.ask_next_view Return ordered NextViewRequest records ranked by expected information gain. Read vision.binary.authorize Validate an AnalysisAuthorization and report exactly what it permits. Read vision.binary.build_call_graph Deterministic call graph: sorted nodes, sorted edges, stable output. Read vision.binary.compare_versions Inventory two authorized bundles and produce a deterministic resource diff. Read vision.binary.decompile_function Decompile one named function. Approximate evidence, never source. Read vision.binary.functions Normalize bounded function observations without claiming source authority. Read vision.binary.identify_format Identify container format and architecture statically, without Ghidra. Read vision.binary.inspect Statically identify and hash one explicitly authorized regular file. Read vision.binary.list_cross_references References into each recovered function entry point. Read vision.binary.list_exports External entry points the binary exports. Read vision.binary.list_functions Functions Ghidra recovered, with sizes and outgoing call edges. Read vision.binary.list_imports External symbols the binary imports, with their libraries. Read vision.binary.list_resources Bounded resource inventory of a bundle. Nothing is extracted. Read vision.binary.list_strings Defined strings with addresses. Untrusted data, never instructions. Read vision.binary.list_symbols Every symbol in the recovered symbol table. Read vision.binary.strings Return capped printable strings as untrusted data, never instructions. Read vision.binary.verify Re-derive the analysis and report whether it reproduces bit for bit. Read vision.build_world_model vision.build_world_model Read vision.bundle.graph Inventory a directory or ZIP-compatible bundle without extraction. Read vision.calibrate_sensor Calibrate lens/principal-point from checkerboard images. Read vision.capabilities List available, blocked, and experimental capabilities for the active profile. Read vision.capture_state Return one actually observed state and its exact screenshot citations. Read vision.capture_world Evaluate coverage and seal a SceneEvidenceGraph with visibility-capped authority. Read vision.close_stream Close an open ocular stream and release capture resources. Read vision.compare Compare two captures by digest identity and overview metadata. Read vision.compare_backends Queue license-aware backend consensus that retains incompatible hypotheses. Read vision.compare_camera_solutions Queue camera consensus while retaining hypotheses with incompatible scale. Read vision.compare_reconstruction_backends Compare reconstruction candidates metric-by-metric without a single fused score. Read vision.compose_camera_path Compose a sealed CameraPathGraph; authority is derive() of declared inputs. Read vision.discover_states Discover only actually observed state, interaction, responsive, and motion evidence. Read vision.establish_correspondence Match target and candidate regions from semantics, text, layout, IDs, and visual features. Read vision.explain_belief Return the append-only belief history for one entity slot. Read vision.explain_region Explain evidence, authority, and uncertainty at an observed pixel. Read vision.fit_replacement_font Fit redistributable replacement fonts to observed geometry without target font bytes. Read vision.fixate Fixate gaze on a region with inhibition-of-return and attention budget. Read vision.gate_material_parity Gate material parameters only when exact-rig sensitivity and confounder proof exists. Read vision.generate_archetype Instantiate a procedural archetype at PROCEDURAL_GROUND_TRUTH; optional Blender emit. Read vision.generate_procedural_scene Compile a sealed ProceduralSceneGraph at PROCEDURAL_GROUND_TRUTH authority. Read vision.get_artifact Resolve a content-addressed artifact by SHA-256 digest. Read vision.get_stream_state Return buffer occupancy, timestamps, drops, and execution class for a stream. Read vision.index.code_search Search code tokens and declared/parsed symbols. Read vision.index.geometry_search Search bounded mesh bounds and classical geometry signatures. Read vision.index.metadata_search Search flattened source, license, and generic metadata tokens. Read vision.index.search Search classical text-token descriptors. Read vision.index.similar Find cross-format exact and near entries for one indexed artifact. Read vision.index.visual_search Search image dHash, colour, edge, and silhouette descriptors. Read vision.infer_materials Infer a MaterialHypothesisSet capped by derive() over observation authorities. Read vision.inspect_graphics Capture or progressively disclose governed canvas and graphics-runtime evidence. Read vision.list_artifacts List content-addressed artifacts recorded in the project. Read vision.list_surprises List recorded surprise events for a world (optionally filtered). Read vision.list_uncertainties List entities ordered by uncertainty (lowest confidence first). Read vision.measure_information_gain Estimate expected information gain for a proposed view on a target. Read vision.observe Capture a local visual target into durable OBSERVED evidence (no browser). Read vision.observe_change Classify session-to-session world changes (moved/removed/new/lighting). Read vision.open_project Open an existing project path and remember it for this session. Read vision.open_stream Open a calibrated ocular stream (video, sequence, screen, render, or opt-in webcam). Read vision.optimize_inverse_render Joint material/light solve; records stay separate and never share authority. Read vision.plan_capture Propose coverage-maximising views; sealed plan carries HYPOTHETICAL authority. Read vision.plan_parity_repairs Rank bounded repair hypotheses with files, predicted effects, blast radius, and gates. Read vision.predict_next Predict pose/visibility/existence for the next horizon frames. Read vision.progress Return compact perception progress and evidence counts. Read vision.query Query an observed perceptual graph with exact artifact citation. Read vision.query_world Query entities, relations, or a scene summary from a world model. Read vision.reidentify Attempt re-identification of a detection against lost/occluded tracks. Read vision.render.benchmark Report the render benchmark matrix status. Read vision.render.cache_info Show the cache identity for render requests. Read vision.render.cache_prune Describe which cached render classes would be dropped. Read vision.render.draft Prune candidates from draft evidence. Read vision.render.plan Plan a draft-verify render search. Executes nothing. Read vision.render.verify Describe what a full-Cycles verification of this request requires. Read vision.resolve_target Resolve the canonical governed target through the stable vision surface. Read vision.resource.inventory Return bounded resource metadata while preserving rights and no-extract state. Read vision.resource.usage_edges Build an approximation-only binary-to-pixel graph from observations. Read vision.review_queue List items that still require named human or policy review. Read vision.search.compare Explain every score component and eligible winner for two candidates. Read vision.search.deduplicate Group exact candidate identities without deleting evidence. Read vision.search.explain_candidate Explain rights, disqualifiers, score, and publication eligibility. Read vision.search.ledger Read a digest-verified search/candidate/download/index/receipt ledger. Read vision.search.local Search only explicitly authorized local roots with bounded file reads. Read vision.search.models Register 3D-model candidates without downloading unresolved licenses. Read vision.search.query_index Query the classical local content index by text tokens. Read vision.search.rank Apply hard disqualifiers then the independent 100-point model score. Read vision.search.receipt Verify a Greed V2 receipt against database and content-addressed bytes. Read vision.search.repositories Register repository/release leads from a governed source adapter. Read vision.search.resolve_identity Resolve exact and near identities and flag false exact claims. Read vision.search.resolve_license Adjudicate candidate-bound rights from digest-bound license evidence. Read vision.search.resolve_rights Resolve analysis/download/redistribution rights independently. Read vision.search.resolve_version Resolve an exact or nearest ordered version candidate. Read vision.search.versions Build a version/variant/source-format lineage graph. Read vision.search.web Register governed web-search metadata; remote text remains untrusted data. Read vision.solve_layout Infer explainable grid, gutter, rhythm, breakpoint, sticky, and crop constraints. Read vision.solve_pnp_landmarks Recover pending metric cameras from an immutable named landmark review. Read vision.solve_vanishing_points Recover focal length and orientation from x/y/z perspective-grid vanishing points. Read vision.stream_scene_assets Build a scroll-gated streaming plan (poster→shell→detail→prefetch→terminal). Read vision.trace_behavior Trace observed input-to-visual-effect causality for an experience capture. Read vision.validate_3d_semantics Detect GLB semantic-identity loss independently of polygon count. Read vision.verify Verify a capture by evidence digest or verify a receipt file. Read visual_geometry.audit_manufactured_form Audit malformed meshes and manufactured-form consistency from scene evidence. Read visual_geometry.diagnose_residuals Translate replayable residuals into bounded diagnostic repair hypotheses. Read visual_geometry.list List fixed rigs, scorecards, and manufactured-form audit records. Read visual_geometry.score_frequencies Aggregate primary, secondary, and tertiary component scores without area masking. Read visual_geometry.verify Replay and verify a fixed rig, scorecard, or manufactured-form audit receipt. Read worker.claim Claim the best hardware-, load-, model-, and locality-matched queued job. Read worker.heartbeat Refresh worker load, warm-model state, and content-addressed artifact locality. Read worker.list List registered workers, advertised hardware, current load, and liveness. Read worker.renew Renew an authenticated live job lease while a long operation is progressing. Read workflow.prepare_acceptance Refresh coverage and receipt gates, then return the named human-review queue. Read workflow.prepare_capture_session Combine frame triage, coverage, and concrete capture guidance. Read workflow.progress Return compact recomputed progress without changing evidence or acceptance state.
EXPLORE

Related servers

Other MCP servers with similar tools — same risk classification, starter policies for each.

UnClick 1662 tools
adminautomation
Delx Mcp A2a 1082 tools
adminautomation
Delx Protocol — Agent Recovery & Continuity 1076 tools
adminautomation
Nodebench 824 tools
adminautomation
MCP Framework Personal 811 tools
adminautomation
Delx MCP Server 740 tools
adminautomation
Binance MCP Server 734 tools
adminautomation
Eda Agent 726 tools
adminautomation
FAQ

Questions about Visionmcp

Can an AI agent delete data through the Vision MCP server? +

Yes. The Visionmcp server exposes 2 destructive tools including active_learning.rollback, artifact.reap_stale_uploads. These permanently remove resources with no undo. PolicyLayer blocks destructive tools by default so they never reach the upstream server.

How do I prevent bulk modifications through Visionmcp? +

The Visionmcp server has 120 write tools including active_learning.promote, active_learning.record_corrections, artifact.abort_upload. Set a rate limit in your policy -- for example, 10 calls per hour prevents an agent from making more than 10 modifications per hour. PolicyLayer enforces this at the gateway, before calls reach Visionmcp.

How many tools does the Vision MCP server expose? +

376 tools across 4 categories: Destructive, Execute, Read, Write. 186 are read-only. 190 can modify, create, or delete data.

How do I enforce a policy on Visionmcp? +

Register the Vision MCP server in PolicyLayer, apply the suggested rules above (adjust the limits to your use case), and point your AI client at the PolicyLayer proxy URL instead of the server directly. Your agents keep the same tools; PolicyLayer evaluates every call against policy before it executes. Nothing to install, live in minutes.

Enforce policy on every Visionmcp tool call.

Deterministic rules across all 376 Visionmcp tools. Per-identity grants. Full audit log. Live in minutes. Nothing to install.

ENFORCE POLICY ON VISIONMCP →

Instant setup, no code required.

376 Visionmcp tools catalogued and risk-classified — across an index of 46,500+ MCP servers.

// WHERE THIS COMES FROM

These policies come from Vision's registry record.

The record behind this page: verified identity, auth posture, risk grade, every tool classified, recommended policy — re-checked continuously.

Teams ship this data inside their own products. See what a licence covers →

// GET IN TOUCH

Have a question or want to learn more? Send us a message.

Message sent.

We'll get back to you soon.