Use the app's AI gateway: chat, embeddings, list models, read/update config, read usage. Actions: - chat { app_id, messages, model?, temperature?, max_tokens? } Synchronous (no streaming). Returns the full assistant response. Default model is the app's configured default, or "openai/gpt-4o-mini"....
Risk signalsAccepts raw HTML/template content (messages[].content) · High parameter count (26 properties)
Part of the Mcp server.
Free to start. No card required.
AI agents use manage_ai to create or modify resources in Mcp. Write operations carry medium risk because an autonomous agent could trigger bulk unintended modifications. Rate limits prevent a single agent session from making hundreds of changes in rapid succession. Argument validation ensures the agent passes expected values.
Without a policy, an AI agent could call manage_ai repeatedly, creating or modifying resources faster than any human could review. PolicyLayer's rate limiting ensures write operations happen at a controlled pace, and argument validation catches malformed or unexpected inputs before they reach Mcp.
Write tools can modify data. A rate limit prevents runaway bulk operations from AI agents.
{
"version": "1",
"default": "deny",
"tools": {
"manage_ai": {
"limits": [
{
"counter": "manage_ai_rate",
"window": "minute",
"max": 30,
"scope": "grant"
}
]
}
}
}See the full Mcp policy for all 47 tools.
These attack patterns abuse exactly the kind of access manage_ai gives an agent. Each links to the full case and the policy that stops it:
Other write tools across the catalogue. The same approach applies to each: rate-limit and validate the arguments.
Use the app's AI gateway: chat, embeddings, list models, read/update config, read usage. Actions: - chat { app_id, messages, model?, temperature?, max_tokens? } Synchronous (no streaming). Returns the full assistant response. Default model is the app's configured default, or "openai/gpt-4o-mini". - embed { app_id, input (string | string[]), model?, encoding_format? } Returns OpenAI-shaped embedding response. - list_models { app_id } Returns { models: AiModel[] } — discover what the app can call. - get_config { app_id } Returns { defaultModel, allowedModels, maxTokensPerRequest, ... } - update_config { app_id, config } Set defaultModel, allowedModels, maxTokensPerRequest (1–100000), or rotate BYOK. - get_usage { app_id, startDate?, endDate? } Aggregate token counts and costs over a window. - submit_video { app_id, model, prompt, duration?, resolution?, aspect_ratio?, generate_audio?, seed? } Submits an async video-generation job. Returns { job_id, status, polling_url }. Poll the returned URL until status is "completed". - poll_video { app_id, job_id } Returns current { status, model, content_urls?, error?, created_at }. When status is "completed", content_urls contains absolute URLs (same origin as the polling_url) that the caller can fetch() directly using the same Authorization header. Use this to drive your own polling loop. This tool wraps the same /v1/:app_id/chat/completions, /embeddings, /ai/config, /ai/models, /ai/usage routes the SDK uses. The "chat" action sets stream: false; for streamed deltas, drive the SDK from inside a function or DO.. It is categorised as a Write tool in the Mcp MCP Server, which means it can create or modify data. Consider rate limits to prevent runaway writes.
Register the MCP server in PolicyLayer and add a rule for manage_ai: 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 Mcp. Nothing to install.
manage_ai is a Write tool with medium risk. Write tools should be rate-limited to prevent accidental bulk modifications.
Yes. Add a rate_limit block to the manage_ai 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 manage_ai. 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.
manage_ai is provided by the MCP server (@butterbase/mcp). PolicyLayer sits as a proxy in front of this server to enforce policies before tool calls reach the server.
Deterministic rules across all 47 Mcp tools. Per-identity grants. Full audit log. Live in minutes. Nothing to install.
Free to start. No card required.
4,600+ MCP servers and 31,000+ tools scanned and risk-classified.