Use Fainflow from your AI (MCP)

Fainflow exposes an MCP server of its own, so any MCP-capable AI client — Claude, ChatGPT, Cursor, VS Code, n8n and others — can work with your flows, connections and bridges. You sign in with your regular Fainflow account (OAuth); no API keys to copy.

There are two endpoints:

https://mcp.fainflow.com/mcp               # your whole workspace
https://mcp.fainflow.com/mcp/<bridgeId>    # one bridge only (its tools + context)

The workspace endpoint gives the AI tools to list and build flows and bridges, plus every tool federated from your MCP connections. The per-bridge endpoint is a narrow, safe slice — see Bridge.

Claude (web / desktop)

  1. Settings → Connectors → Add custom connector.
  2. URL: https://mcp.fainflow.com/mcp.
  3. Click Connect — a Fainflow sign-in window opens; log in and approve.

Claude then sees your Fainflow tools in every chat where the connector is enabled.

Claude Code (CLI)

Add the server, then authenticate — Claude Code opens your browser and handles the OAuth loopback automatically. It picks a free local port each time and Fainflow accepts any loopback callback, so there's nothing to configure:

claude mcp add --transport http fainflow https://mcp.fainflow.com/mcp

Then run /mcp inside Claude Code and pick fainflow to sign in. You can also add it by hand to .mcp.json:

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

Cursor

Settings → MCP → Add new MCP server, transport HTTP, URL https://mcp.fainflow.com/mcp. Cursor opens the browser for sign-in and returns automatically.

ChatGPT

Settings → Connectors → Advanced → Developer mode, then add a connector with the MCP server URL https://mcp.fainflow.com/mcp and authentication OAuth. Approve the Fainflow sign-in when prompted.

VS Code (GitHub Copilot)

Add to your mcp.json (Command Palette → MCP: Add Server → HTTP):

{
  "servers": {
    "fainflow": { "type": "http", "url": "https://mcp.fainflow.com/mcp" }
  }
}

VS Code walks you through the OAuth sign-in on first use.

n8n

Use the MCP Client Tool node. Fainflow's endpoint speaks Streamable HTTP, which the node supports from n8n 1.104+ (earlier versions only offer the deprecated SSE transport):

  1. Add MCP Client Tool to your workflow (as an AI Agent tool).
  2. Endpoint: https://mcp.fainflow.com/mcp (or a per-bridge URL).
  3. Server Transport: HTTP Streamable. Authentication: OAuth2.
  4. Complete the Fainflow sign-in when n8n opens it.

Self-hosted n8n: OAuth callbacks must be pre-registered. n8n Cloud works out of the box; for a self-hosted instance contact us with your instance URL, or connect a per-bridge endpoint through a bearer-style proxy.

Other MCP clients (mcp-remote)

Clients that only speak stdio can use mcp-remote:

{
  "mcpServers": {
    "fainflow": {
      "command": "npx",
      "args": ["mcp-remote", "https://mcp.fainflow.com/mcp"]
    }
  }
}

What the AI can do

After connecting the workspace endpoint, the client's tool list includes:

  • whoami, list_flows, get_flow, create_flow, set_flow_data, list_flow_templates — inspect and author flows
  • create_api_trigger_route, bind_flow_route, enable_flow — take an API-trigger flow all the way live (same plan-limit and graph validation as the dashboard)
  • list_flow_history, get_flow_execution — check whether a flow you built actually ran (execution values are stripped for privacy); delete_flow_history, delete_flow to clean up
  • create_api_key, list_api_keys, delete_api_key — manage the keys that call an API-trigger flow (the key is shown once)
  • set_flow_secret, list_flow_secrets, delete_flow_secret — manage the per-flow vault a flow references as {{vault.KEY}}
  • list_connections, describe_connection_tools — see what's connected
  • list_bridges, get_bridge, create_bridge, set_bridge_graph, set_bridge_sources, delete_bridge — build bridges
  • every tool from your connected MCP servers (GitHub, Atlassian, …), prefixed per connection

Everything runs as you: the AI only sees your own flows, connections and bridges, and every action is covered by the same audit logging as the dashboard.

Troubleshooting

  • 401 after connecting — the token expired and the client didn't refresh; disconnect and reconnect the server in the client.
  • Sign-in window never returns (CLI clients) — the client's browser tab didn't hand the callback back; close it, re-run /mcp (Claude Code) and complete the sign-in in the fresh tab. Fainflow accepts the loopback callback on any port, so there's no port to set.
  • Empty tool list beyond the basics — you have no MCP connections yet; add one in the dashboard first (how).