Compare two images pixel-by-pixel and identify visual differences. Returns structured diff data including: - Overall diff percentage and pixel counts - Clusters of different regions with bounding box coordinates (x, y, width, height) - Severity rating per cluster (trivial/minor/moderate/major) - ...
AI agents call get_diff_of_images to retrieve information from Langfuse Observability without modifying anything — typically the context-gathering step in research, monitoring, and reporting workflows, before the agent takes action elsewhere.
This tool only reads and compares two images to produce analytical output (diff percentages, pixel counts, bounding boxes, heatmaps). It has no side effects, does not modify any state, and does not execute arbitrary code or trigger external operations. The output is descriptive metadata about visual differences, making this a straightforward Read operation with minimal security risk.
From the tool's definition Tool performs image comparison and analysis without modifying data. Description states it 'Compare[s] two images pixel-by-pixel and identify[s] visual differences' and 'Returns structured diff data' — pure retrieval/analysis operations.
Attacks that exploit this kind of access
Compare two images pixel-by-pixel and identify visual differences. Returns structured diff data including: - Overall diff percentage and pixel counts - Clusters of different regions with bounding box coordinates (x, y, width, height) - Severity rating per cluster (trivial/minor/moderate/major) - Clustering metadata with the gap used and suggestions for tuning - File paths to generated heatmap images showing diff intensity Clustering: By default, nearby diff regions are automatically grouped using a natural-breaks algorithm that finds the optimal merge distance. This produces a manageable number of clusters where each represents a distinct problem area. To override, set cluster_gap explicitly (0 = no merging, or a specific pixel distance). Heatmap output: A PNG image where diff regions are colored from yellow (subtle difference) to red (major difference). A composite version overlays this on the source image for easy comparison. Use cases: - Compare a design mock screenshot against actual UI implementation - Compare a Figma mock of a single component against a full-page screenshot (auto-aligned) - Detect visual regressions between two versions of a page - Verify CSS/layout changes only affect intended areas Requirements: - Images must be accessible as local file paths - Images can have different dimensions — the smaller image is automatically aligned within the larger one using template matching Note: Anti-aliased pixels (font smoothing, rounded corners) are excluded by default to reduce false positives. Set include_aa=true to include them. It is categorised as a Read tool in the Langfuse Observability MCP Server, which means it retrieves data without modifying state.
Register the Langfuse Observability MCP server in PolicyLayer and add a rule for get_diff_of_images: allow, deny, rate-limit, or require approval. Point your MCP client at the PolicyLayer proxy URL and the rule is enforced on every call, before it reaches Langfuse Observability. Nothing to install.
get_diff_of_images is a Read tool with low risk. Read-only tools are generally safe to allow by default.
Yes. Add a rate_limit block to the get_diff_of_images rule in your PolicyLayer policy. For example, setting max: 10 and window: 60 limits the tool to 10 calls per minute. Rate limits are tracked per agent session and reset automatically.
Set action: deny in the PolicyLayer policy for get_diff_of_images. The AI agent will receive a policy violation error and cannot call the tool. You can also include a reason field to explain why the tool is blocked.
get_diff_of_images is provided by the Langfuse Observability MCP server (langfuse-observability-mcp-server). PolicyLayer sits as a proxy in front of this server to enforce policies before tool calls reach the server.
Every MCP server has a record like this.
Type a name, get the same breakdown: verified identity, auth posture, risk grade, capabilities, recommended policy.
Teams ship this data inside their own products. See what a licence covers →