mcp tool drift check

    1

    Third-party Model Context Protocol (MCP) servers often update their tool manifests without notice.

    Free

    1 useSecurity scanned

    Works with the AI Agents you already use

    Claude CodeClaude CodeCursorCursorCodex CLICodex CLIMuseMuseDotsDots+21 more

    mcp tool drift check

    Example session with this skill installed

    Check if the MCP server at https://mcp.brave.com/search has any tool drift.

    • Read your context and instructions
    • Compiled the mcp tool drift

    Live response from the Dominion Observatory free drift check (run just now):

    {
      "server": "https://mcp.brave.com/search",
      "state": "NO_BASELINE",
      "supported": null,
      "tracked": false,
      "baseline": null,
      "drift_detected": null,
      "next_step": {
        "method": "POST",
        "url": "https://dominionobservatory.com/api/drift/baseline",
        "note": "Capture publishes the manifest and compares against the previous successful capture. First capture is not detected drift."
      }
    }
    

    What this tells the agent

    FieldValueMeaning
    stateNO_BASELINEFirst observation of this endpoint - there is no earlier snapshot to compare against yet.
    drift_detectednullNo verdict possible on a first check. This is expected, not an error.
    supportednullCompatibility is determined when a capture runs.

    Next step for the agent: record what the manifest looks like now, then re-run the same check after the server's next release. A later drift_detected: true means the tool names, descriptions or input schemas changed since the baseline - review the diff before letting your agent call the tools again.

    For scheduled checks with email and webhook alerts instead of manual re-checks, the skill points to the US$29/server/month monitoring plan at https://dominionobservatory.com/monitor.

    Connects securely to your tools. The creator never sees your data.

    About this skill

    Third-party Model Context Protocol (MCP) servers often update their tool manifests without notice. When tool names, descriptions, or input schemas change, your agent's pre-built calls break or hallucinate. This skill verifies the current state of any public MCP endpoint against known baselines to detect silent schema drift.

    What it does

    • Endpoint validation checks if a public MCP URL is compatible with standard list tools requests.
    • Drift detection compares the live manifest against a recorded baseline to identify changes.
    • Manifest snapshotting records a cryptographically signed state of the server for future comparison.
    • Compatibility reporting identifies if an endpoint supports SSE or JSON responses over HTTPS.
    • Verification allows agents to validate Ed25519-signed captures against a public JWKS.

    How it works

    1. Locate the target MCP server's public endpoint URL.
    2. Execute a drift check via the Dominion Observatory API.
    3. Parse the JSON response to determine the state and drift_detected flag.
    4. Review the baseline metadata if a previous snapshot exists to see what changed.

    Frameworks & tools

    This skill interacts with the Dominion Observatory API using standard HTTPS GET requests and handles JSON payloads. It is compatible with any agent environment capable of making web requests, such as Claude Code or Cursor.

    Why this beats prompting it yourself

    Prompting an agent to "check for changes" usually results in it just re-reading the current docs without a historical reference. This skill provides access to a persistent, signed version history that the agent cannot maintain on its own.

    Use cases

    • Verify MCP tool schemas before executing high-stakes autonomous actions.
    • Debug integration failures when an MCP server starts returning validation errors.
    • Monitor public MCP endpoints for breaking API changes during development.
    • Audit the integrity of third-party tool definitions using signed captures.

    Known limitations

    Does not support authenticated endpoints, redirects, or IPv6-only hosts. Only works with unauthenticated JSON or SSE responses over HTTPS. Free checks are one-time and manual.

    No support for auth, redirects, pagination, or IPv6. Only checks public HTTPS endpoints. Free tier is manual; automated monitoring requires a paid subscription.

    How to use

    1. 1

      Connect your agent to the Agensi MCP

      Copy the prompt and paste it into Claude, ChatGPT, Cursor or any agent that supports MCP. It sets up the connection for you. Manual setup

    2. 2

      Sign in to Agensi

      Your agent shows a sign-in link. Sign in and click Allow.

    3. 3

      Ask for what you need

      Your agent loads this skill and uses it on your task. It always asks before any payment.

    Reviews

    No reviews yet

    Be one of the first to try it. Every listed skill goes through the trust checks below.

    Security scanned

    Checked by our 8-point scan before listing

    1 use

    Times used to date

    Free forever

    No account required to browse

    Trust & safety

    Security scan

    Security scan passed 1 month ago

    • Free with an Agensi account

    Permissions

    Network Access
    Listed1 month ago

    What's inside

    Frequently Asked Questions

    This skill acts as an automated 'early warning system' that compares live MCP manifests against a cryptographically signed baseline to detect changes in tool names, descriptions, or input schemas that cause agents to fail.

    It works with any environment capable of making HTTPS GET requests, including Claude Code, Cursor, and custom agent frameworks that support Model Context Protocol integrations.

    No, this skill is designed specifically for public, unauthenticated HTTPS endpoints (JSON or SSE) and does not currently support authenticated servers, redirects, or IPv6-only hosts.

    Installation is straightforward: you connect your agent to the Dominion Observatory API via the provided endpoint and use the drift check tool to begin snapshotting your required MCP manifests.

    Unlike simple prompting, this skill provides a persistent, signed version history and a 'baseline' reference that allows the agent to perform objective technical comparisons rather than relying on its own memory.

    New in LLM Ops