More screenshots

    Works with the AI Agents you already use

    Claude CodeClaude CodeCursorCursorCodex CLICodex CLIDeepSeekDeepSeekMuseMuse+22 more

    MCP Drift

    1

    Detect breaking changes and schema drift in MCP server tool definitions with deterministic contract testing.

    Free

    Security scanned

    MCP Drift

    Example session with this skill installed

    We just upgraded the MCP server named "demo" from 1.0.0 to 2.0.0. Check it against our approved baseline and tell me if anything will break our agents.

    • Read your context and instructions
    • Compiled the mcp drift

    Summary

    I ran mcp-drift check --all against your approved baseline. The upgrade is

    not safe to merge yet: 2 breaking changes, 2 changes to review, 2 informational. Exit code 1.

    Server demo: 1.0.0 → 2.0.0 · protocol 2026-07-28 · tools 3 → 3

    Findings

    CategoryToolRulePathBeforeAfter
    🔴 breakingread_filetool.removed/advertised(absent)
    🔴 breakingsearch_filesinput.required_added/inputSchema/required["query"]["folder","query"]
    🟡 reviewsearch_filestool.description_changed/description"Search file names in the workspace.""Search file names inside one folder of the workspace."
    🟡 reviewwrite_notetool.added/(absent)new tool, advertises destructiveHint: true
    🔵 infolist_foldersinput.constraint_changed/inputSchema/properties/max_depth/maximum310
    🔵 infolist_folderstool.title_changed/title(absent)"List folders"

    What this means

    read_file was removed. Any agent that calls it can fail. A rename also shows up as a removal plus an addition, so check whether it was replaced.

    search_files now requires folder. Existing requests that only send query can be rejected. This is a stricter input, which is breaking for callers.

    The search_files description changed. Agents read descriptions to decide when to call a tool. This is an advertised change, not proof the behaviour changed.

    write_note is new and advertises a destructive hint. Nothing existing breaks, but a new capability is now available to your agents. Review what it does.

    • list_folders allows a larger max_depth (3 → 10). Looser input, so previously valid requests stay valid.

    Reports saved to artifacts/mcp-drift.md and artifacts/mcp-drift.json.

    Next steps

    1. Update the callers of read_file and search_files, or pin the server back to 1.0.0.
    2. Decide whether your agents should be allowed to use write_note.
    3. When the changes are intended and reviewed, tell me and I will run:
      mcp-drift snapshot --all --out .mcp-drift/candidate.json
      mcp-drift baseline accept .mcp-drift/candidate.json --replace

    I have not changed your baseline. Accepting it is your decision.

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

    About this skill

    MCP servers publish tool definitions that can change silently during upgrades, leading to broken integrations when a tool disappears or an input becomes required. This skill manages tool-contract drift by maintaining a verified baseline of your MCP servers and comparing it against live definitions. It provides deterministic, code-based analysis to ensure your agentic workflows remain stable without manually checking JSON schemas.

    What it does

    • Baseline management captures a reviewed snapshot of all tool definitions including schemas and annotations.
    • Breaking change detection identifies stricter input requirements or looser output rules that could break callers.
    • Multi-transport support connects to servers over both local stdio and remote Streamable HTTP.
    • CI automation generates machine-readable JSON and human-readable Markdown reports for GitHub Actions.
    • Policy enforcement triggers specific exit codes based on finding categories like breaking, review, or info.

    How it works

    1. Initialize the project configuration and define server commands or URLs in mcp-drift.toml.
    2. Snapshot current tool definitions to create a candidate baseline for human review.
    3. Accept the baseline after verifying tool details, creating a source of truth for the repository.
    4. Check for drift regularly to identify schema changes, added tools, or removed capabilities.

    Frameworks & tools

    This skill utilizes the mcp-drift CLI, Python 3.11+, and integrates with GitHub Actions for automated contract testing. It supports standard MCP transports including stdio (via npx or local commands) and Streamable HTTP.

    Why this beats prompting it yourself

    Manual schema comparison is error-prone and misses subtle breaking changes like narrowing an enum or changing a required field. This skill provides a deterministic, repeatable workflow that functions in CI pipelines where LLMs are too slow or non-deterministic for contract validation.

    Use cases

    • Prevent production failures when upgrading third-party MCP servers.
    • Audit internal MCP server changes before merging PRs.
    • Document tool capabilities across a distributed team of agent developers.
    • Enforce strict interface contracts in multi-agent systems.

    Known limitations

    Focuses on tool contracts only, not resources or prompts. Analysis of complex JSON Schema features like allOf or oneOf is flagged for manual review rather than automatically classified. Tested on Linux; macOS and Windows support is not yet validated.

    Find more skills like this on the Claude Skills Marketplace, the Codex Skills Marketplace and the Gemini Skills Marketplace, or browse Claude Skills for LLM Ops.

    Only tracks tool schemas, not resources or prompts. OS support currently limited to Linux. Requires Python 3.11+. Complex JSON Schema features are flagged for review rather than auto-resolved.

    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 buying a skill.

    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

    Fresh listing

    Recently published to Agensi

    Free to use

    No account required to browse

    Trust & safety

    Security scan

    Security scan passed today

    • Free with an Agensi account

    Software
    PythonGit

    Permissions

    Terminal / Shell
    Network Access
    Read Files
    Write Files
    Environment Variables
    Listedtoday

    What's inside

    Frequently Asked Questions

    MCP Drift provides deterministic contract testing to detect breaking changes in tool definitions, preventing agentic workflows from failing when an underlying MCP server updates its schema without notice.

    The skill supports any MCP-compliant server using stdio (via npx or local commands) or Streamable HTTP transports, making it compatible with most standard MCP implementations.

    No, this skill uses code-based analysis to identify schema drift, providing a faster, more reliable, and lower-cost alternative to using LLMs for contract validation in CI/CD pipelines.

    It specifically identifies stricter input requirements, changed enum values, and removed tools, generating both machine-readable JSON for automation and Markdown reports for human review.

    You receive the mcp-drift CLI tool, documentation for GitHub Actions integration, and the configuration framework needed to establish and maintain tool baselines.

    Currently, the skill focuses on tool contracts and schemas; it does not analyze changes to MCP resources or prompts at this time.

    New in LLM Ops

    Free