MCP Charts

What MCP Charts does

MCP Charts MCP server provides an agent-facing directory for discovering MCP servers and other reusable AI resources. The catalog covers six resource types: MCP servers, skills, plugins, agents, prompts, and rules. Results can help an agent narrow a task to candidate resources, inspect a catalog entry, and follow the upstream setup instructions.

The service is hosted and read-only. Its tools do not install, execute, or modify the resources they return. Use the catalog for discovery and evaluation, then handle installation or configuration through the upstream project.

How it works

MCP Charts MCP server is available at https://mcpcharts.com/mcp over Streamable HTTP. It does not require authentication, an account, or an API key. The catalog is refreshed continuously from public sources, including the official MCP Registry.

Client compatibility is based on manifests and registry declarations rather than runtime testing. Repository stars and activity describe the source repositories, not individual catalog entries. Requests are rate-limited, and result sets have fixed bounds, so broad searches may require multiple pages or narrower terms.

Setup and configuration

Because the service is hosted, there is no local package to install or process to run. Add the endpoint as a remote MCP server in a compatible client. For Claude Code, register the HTTP endpoint with the client’s MCP command. For clients using mcp.json, add a server named mcpcharts with this URL:

{
  "mcpServers": {
    "mcpcharts": {
      "url": "https://mcpcharts.com/mcp"
    }
  }
}

Some clients represent the same connection with a separate type or transport field for Streamable HTTP. Follow the client’s remote-server configuration format. The endpoint can also be added as a custom connector in Claude.ai or ChatGPT.

Tools and capabilities

MCP Charts MCP server exposes three read-only tools:

• recommend_resources ranks a shortlist for a task. It accepts a required task and optional client, type, and limit values. The limit ranges from 1 to 8, and keyword relevance is prioritized ahead of repository stars.
• search_resources searches keywords across the catalog. It accepts query, with optional type and page parameters, and returns up to 24 results per page.
• get_resource looks up one catalog entry by id. Its response includes compatibility evidence, freshness information, and the upstream setup link.

The type filter accepts skill, mcp, plugin, agent, prompt, or rule. Example tasks include finding a Postgres MCP server for Claude Code, locating skills for Playwright tests, or selecting a pull-request review plugin for Cursor.

Limitations and notes

Catalog compatibility is evidence-based rather than verified through runtime tests. Before using a result, check the upstream project’s requirements, especially constraints such as free usage or local-only operation. The service only provides discovery and reference information; it does not install or run returned resources.