MyMCPTools

npx add-mcp 'https://mymcptools.com/api/mcp'

MyMCPTools is an MCP server

The directory answers over the Model Context Protocol itself. Point any MCP client at the endpoint below and your agent can search 2,506 MCP servers and check whether one is actually reachable — from 2,440 live-probed status records — before it recommends an install.

Endpoint

https://mymcptools.com/api/mcp

Transport

Streamable HTTP

Authentication

None — open and read-only

Sessions

Stateless (POST only)

Connect

Claude Code — one command:

claude mcp add --transport http mymcptools https://mymcptools.com/api/mcp

Claude Desktop, Cursor, Windsurf and other config-file clients — add this to your MCP config:

{
  "mcpServers": {
    "mymcptools": {
      "type": "http",
      "url": "https://mymcptools.com/api/mcp"
    }
  }
}

Claude connector (claude.ai)

The connector listed in Claude’s directory uses its own URL with the same eight read-only tools and data. No account, API key or sign-in is needed. To add it by hand, open Claude → Customize → Connectors → Add custom connector and paste:

https://mymcptools.com/api/mcp/claude

Things to try: “Find an MCP server for Postgres”, “Has the GitHub MCP server had incidents recently?”, “Which remote MCP servers for Slack answered a live handshake today?”. Results come from our catalog and our own probes; a server that only runs locally (stdio) cannot be probed and shows a repository-freshness signal instead. We do not store your conversation or your search terms — see the privacy policy. Questions or problems: support.

Tools

search_mcp_servers

Search the catalog by keyword, category, client integration or install type. Set only_verified to return just the servers that answered a live handshake.

get_mcp_server

Full entry for one server: install command, repo, supported clients, current verdict, repo-freshness signal and related servers.

get_server_status

Current probe verdict, tool count, handshake latency and negotiated protocol version. Omit the slug for the catalog-wide rollup.

get_server_history

Trailing probe history plus a daily uptime sparkline — the same series behind the uptime badges on this site.

list_server_incidents

Reconstructed outage windows with start, end, duration and failure reason. Filter to ongoing outages only.

list_schema_drift

Tool-schema and protocol-version changes detected between probes — the early warning that a dependency just broke.

list_categories

Every category and client integration with server counts, for driving the search filters.

get_catalog_stats

Aggregate health of the whole MCP population: verdict breakdown, serving share, transport mix, latency percentiles.

Catalog entries are also exposed as MCP resources at mymcptools://server/{slug}, alongside mymcptools://catalog/categories and mymcptools://catalog/stats.

Call it directly

The endpoint is plain JSON-RPC over HTTP POST, so you can drive it without an MCP client:

curl -s -X POST https://mymcptools.com/api/mcp \
  -H 'Content-Type: application/json' \
  -H 'Accept: application/json, text/event-stream' \
  -d '{
    "jsonrpc": "2.0",
    "id": 1,
    "method": "tools/call",
    "params": {
      "name": "search_mcp_servers",
      "arguments": { "query": "postgres", "only_verified": true, "limit": 5 }
    }
  }'

GET returns 405 by design — the endpoint is stateless and offers no server-initiated SSE stream, so there is nothing to subscribe to.

MCP server FAQ

An MCP server is a program that exposes tools, resources and prompts to an AI client over the Model Context Protocol. The client (Claude Desktop, Claude Code, Cursor, VS Code, Cline and others) discovers what the server offers with a tools/list call and then invokes those tools with tools/call. Servers run either locally over stdio, launched by the client as a subprocess, or remotely over Streamable HTTP at a URL like the one on this page.

It turns this directory into something an agent can query mid-task. Instead of a human browsing the catalog, the model calls search_mcp_servers to find candidates, get_mcp_server for the install command and repo, and get_server_status to confirm the server actually answered a live handshake before it recommends installing it.

No. The endpoint is open and read-only, with no authentication and no session state — every call is a plain HTTP POST. The keyed REST API on the developers page is a separate product for the same uptime, incident and drift data.

Run claude mcp add --transport http mymcptools https://mymcptools.com/api/mcp. For config-file clients such as Claude Desktop, Cursor and Windsurf, add an entry under mcpServers with type set to http and url set to the same endpoint.

The server is stateless and POST-only. It opens no server-initiated SSE stream, so there is no long-lived channel for a GET to subscribe to. Send JSON-RPC over POST with an Accept header of application/json, text/event-stream.

Most listings are a snapshot of a README. This catalog is continuously probed, so the same tools that return an entry also return its current verdict, handshake latency, negotiated protocol version, reconstructed outage windows and any tool-schema drift detected between probes — the signals that tell you whether a server is still worth depending on.

Prefer REST?

The same uptime, incident and drift data is available as a keyed JSON/CSV API. See the Status API docs for endpoints and access, or the live status page for the human view.