Skip to content
Mig ration F low
Agents Submit GitHub ↗

Use MigFlow with
your agents.

16 playbooks exposed through three channels so agents and scripts can consume them programmatically — no scraping required.

recommended

MCP — Model Context Protocol

Native integration for Claude Code, Claude Desktop, Cursor, Windsurf, and any MCP-compatible client. The agent can list, fetch, and search playbooks as first-class tools.

Endpoint

__ORIGIN__/api/mcp

Available tools

  • list_playbooks(category?, tag?) — list all playbooks, optionally filtered.
  • get_playbook(slug) — fetch a single playbook with full markdown body and metadata.
  • search_playbooks(query) — free-text search across titles, descriptions, tags, and bodies.

Claude Desktop config 

{
  "mcpServers": {
    "migflow": {
      "url": "$__ORIGIN__/api/mcp"
    }
  }
}

Test with the MCP Inspector 

npx @modelcontextprotocol/inspector $__ORIGIN__/api/mcp

JSON API

For scripts, CI pipelines, and non-MCP automation. Plain HTTP, no protocol overhead.

Endpoints

  • GET /playbooks.json — full list with metadata for every playbook.
  • GET /playbooks/<slug>.json — single playbook with raw markdown body and pre-formatted prompts (Claude / Gemini).

Example

curl $__ORIGIN__/playbooks.json | jq '.playbooks[].slug'

curl $__ORIGIN__/playbooks/enzyme-to-rtl.json | jq '.prompts.claude'

llms.txt

Follows the llmstxt.org convention. Use it for RAG indexers, ChatGPT context loaders, or any tool that ingests structured site maps.

Endpoint

GET /llms.txt — index of all playbooks grouped by category, each linked to its JSON endpoint.

Example

curl $__ORIGIN__/llms.txt

All three channels share the same content. Pick the one that best fits your client.