mcpkit

    1

    Find the problems that break an MCP server in specific MCP clients (Claude, ChatGPT, Cursor, VS Code, Codex and more), with sourced client facts, SDK upgrade help and CI output.

    Free

    2 usesSecurity scanned

    Works with the AI Agents you already use

    Claude CodeClaude CodeCursorCursorCodex CLICodex CLIMuseMuseDotsDots+21 more

    mcpkit

    Example session with this skill installed

    Run mcpkit on my local Node.js server at dist/index.js and check compatibility for Claude and Cursor.

    • Read your context and instructions
    • Compiled the mcpkit

    I have completed the compatibility check for your MCP server. The validator identified one critical naming issue and a missing title annotation that affects how tools are displayed in Cursor.

    MCP Compatibility Report

    ClientCheck IDSeverityMessageFix
    CursorTOOL_NAME_CLIENT_COLLISION❌ FAILfetch.user and fetch_user collide after normalization.Use unique names with only _ or -.
    ClaudeTOOL_TITLE_MISSINGℹ️ INFOTool get_weather is missing a human-readable title.Add title: "Get Weather" to tool metadata.
    GeneralSDK_OUTDATED⚠️ WARNUsing @modelcontextprotocol/sdk@0.5.0.Run npx -y @vrajved/mcpkit upgrade.
    GeneralTRANSPORT_STDOUT_POLLUTION✅ PASSNo illegal stdout writes detected.N/A

    Next steps

    • Rename fetch.user to fetch_user_data in src/tools.ts to avoid the Cursor collision.
    • Add annotations: { title: "..." } to all tool definitions to improve UI rendering.
    • Execute npx -y @vrajved/mcpkit upgrade --apply to move to the latest MCP SDK.

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

    About this skill

    Building Model Context Protocol (MCP) servers is difficult because every client—Claude Desktop, ChatGPT, Cursor, VS Code, and others—implements the protocol with unique constraints. A server that works in one environment might silently fail in another due to name truncation, schema depth limits, or handling of structured content.

    What it does

    • Multi-client validation checks server definitions against known constraints for Claude Desktop, ChatGPT, Cursor, VS Code, Cline, and more.
    • Protocol inspection records exactly what the server exposes including tools, resources, and prompts.
    • Schema auditing identifies deep nesting or root-level combinators that break specific LLM clients.
    • Runtime probing optionally executes tools with readOnlyHint to verify output schema compliance.
    • SDK management detects outdated MCP SDKs across Node.js, Python, Go, and Rust and provides precise upgrade commands.
    • Transport verification catches stdout pollution that corrupts JSON-RPC messages on stdio transports.

    How it works

    1. Run npx -y @vrajved/mcpkit -- [start command] to boot the server and perform a full suite of cross-client checks.
    2. Review the Markdown report for FAIL and WARN statuses, each containing a specific checkId and a concrete fix.
    3. Use the upgrade command to check for SDK updates and optionally apply them if the environment allows.
    4. Integrate with CI using the -f github flag to surface protocol violations directly in pull request annotations.

    Frameworks & tools

    Works with any MCP server regardless of language (Node.js, Python, Go, Rust). Requires Node.js 20+ to run the validator. Supports both stdio and Streamable HTTP transports.

    Why this beats prompting it yourself

    LLMs often guess at protocol limits or rely on outdated documentation. This tool uses a sourced database of real-world client behaviors to catch silent failures, like name collisions in VS Code or schema stripping in Codex CLI, that simple prompting cannot detect.

    Proven on real projects

    opsmcp-mcpkit-audit.png

    Run against a live 4-tool MCP server (an LP-solver exposed over MCP) across all 12 supported clients: 0 fails, 0 warnings, 3 informational notes across the whole surface. This is the same validator you'd run against your own server, not a curated example.

    Use cases

    • Validating a new MCP server before deployment to ensure cross-client compatibility.
    • Debugging why tools disappear in specific editors like Cursor or VS Code.
    • Verifying that tool results containing structuredContent also provide required text fallbacks.
    • Auditing OAuth 2.0 implementations for PKCE S256 compliance required by ChatGPT.

    Known limitations

    The tool requires a functional build of the server to run. Probing is limited to tools marked as read-only to prevent side effects. Remote servers require reachable URLs and valid headers.

    Requires Node.js 20+. The target MCP server can be in any language. The server must be buildable and startable locally or reachable via URL. Tool probing only occurs for tools explicitly declaring a read-only hint.

    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

    2 uses

    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

    Requires

    Use MCP
    @vrajved/mcpkit
    Node.js

    Permissions

    Terminal / Shell
    Network Access
    Read Files
    Environment Variables

    mcpkit spawns the target MCP server as a subprocess to connect and inspect it (Terminal/Shell), and makes network requests to check remote (--url) servers, OAuth metadata endpoints, and the latest SDK version from npm/PyPI (Network Access). It reads the server project's package.json, lockfiles, and virtualenv to detect the installed MCP SDK version (Read Files), and supports passing environment variables through to the spawned server via `-e KEY=VAL` (Environment Variables). File writes only happen if you explicitly run `upgrade --apply`, which updates version pins in requirements files — the default check command never writes anything.

    Listed1 month ago

    What's inside

    Frequently Asked Questions

    mcpkit solves the "works on my machine" problem for MCP servers by validating your server against the specific, often undocumented constraints of clients like Claude Desktop, Cursor, and ChatGPT simultaneously.

    It is language-agnostic and works with any server built in Node.js, Python, Go, or Rust, provided you have Node.js 20+ installed to run the validator itself.

    Yes, you can use the `-f github` flag to generate annotations directly in your Pull Requests, flagging protocol-breaking changes before they reach your production branch. Gate merges on mcpkit's exit code in your branch protection rules to enforce it.

    The tool includes a specialized `upgrade` command that scans your environment for outdated SDKs and provides the exact terminal commands needed to bring your server up to the latest protocol version.

    The official inspector is built for manually exploring and debugging one server at a time. mcpkit is built for automated, unattended compatibility checking across many clients at once, with a sourced database of client-specific quirks like schema depth limits and name truncation rules, so you can run it in CI without a human clicking through a UI.

    To prevent accidental data modification, runtime probing is strictly limited to tools marked with a `readOnlyHint`, ensuring your production data remains safe during the audit.

    Yes, two ways. For a quick check, pass a static header: mcpkit --url https://example.com/mcp -H "Authorization: Bearer ***". For full OAuth-protected servers, run mcpkit oauth login --url https://example.com/mcp once to authenticate interactively, then mcpkit check --oauth --url https://example.com/mcp to run checks using the stored token. Tokens are saved locally per server and refreshed automatically; mcpkit never registers a new client or opens a browser unless you explicitly run oauth login.

    New in LLM Ops