We built an MCP server for OpFlow
Point Claude — or any MCP client — at OpFlow and it can list, read, run, and manage your bots. Here's how, and why.
The Model Context Protocol gives an AI agent a typed set of tools it can call. We shipped an OpFlow MCP server so the automations you build on the canvas are also reachable from an agent — no scraping, no brittle UI clicks, a real API surface designed for models.
What it exposes
The server lives at /api/public/mcp, speaks JSON-RPC 2.0, and authenticates with a bearer token (an opflow_ key you mint in the admin console, stored hashed). It surfaces 11 tools covering the bot lifecycle.
- List and read bots, nodes, and connections.
- Create, update, and validate a bot graph.
- Trigger a bot and read back the run, step by step.
- Inspect run history, payloads, and errors for debugging.
If a human operator can do it on the canvas, an agent should be able to do it through a tool call. That's the bar.
Why it matters
Automation and agents are converging. Deterministic bots are great at the parts you can specify; agents are great at the parts you can't. With the MCP server, an agent can reach into OpFlow, fire a well-tested bot for the deterministic work, and read the result — you get the reliability of a real engine with the flexibility of a model on top.
Add the server to your MCP client, drop in a token, and your OpFlow workspace becomes something an agent can operate. Your whole connector library, now one tool call away.