mcp-agentic-patterns

An MCP (Model Context Protocol) server that exposes the Agentic Patterns Catalog — patterns, compositions (recipes + frameworks), methodologies, anti-patterns and code examples — as resources and tools for AI coding agents like Claude Code, Cursor, Cline and Claude Desktop.

Catalog data is sourced from .

Run it locally (today)

git clone https://github.com/agentpatternscatalog/mcp.git
cd mcp
uv sync                      # or: python -m venv .venv && pip install -e .
uv run mcp-agentic-patterns  # stdio transport (default)

The first run downloads the catalog tarball from github.com/agentpatternscatalog/patterns into ~/.cache/mcp-agentic-patterns/patterns-main/ and reuses it on subsequent runs. No network needed after the first start.

Refresh the cached catalog:

uv run mcp-agentic-patterns --refresh-catalog
# or:  MCP_CATALOG_REFRESH=1 uv run mcp-agentic-patterns

Point at a local checkout instead (skips the network entirely):

uv run mcp-agentic-patterns --catalog-dir /path/to/agent-patterns-catalog
# or:  CATALOG_DIR=/path/to/agent-patterns-catalog uv run mcp-agentic-patterns

Resolution priority: --catalog-dir arg → CATALOG_DIR env → sibling ../agent-patterns-catalog/ checkout → on-disk cache → fresh GitHub fetch → bundled package data.

Wire it into your MCP client

Claude Desktop, Claude Code, Cursor, Cline (and other MCP clients) all read a mcpServers config block. Point them at the local checkout:

{
  "mcpServers": {
    "agentic-patterns": {
      "command": "uv",
      "args": [
        "--directory", "/abs/path/to/mcp",
        "run", "mcp-agentic-patterns"
      ]
    }
  }
}

Or, after pip install -e .:

{
  "mcpServers": {
    "agentic-patterns": {
      "command": "mcp-agentic-patterns"
    }
  }
}

HTTP transport

For non-stdio MCP clients, run the server over streamable-HTTP:

uv run mcp-agentic-patterns http --host 0.0.0.0 --port 8080

Tools

Resources