MCP Server

Why this exists

Once your team starts using an AI assistant day to day — Claude, Cursor, an internal agent — the assistant still has to guess at what your workflow does, because it has no way to see it. It can read code you paste in, but it can't open the actual graph, check whether a past run failed, or tell you a flow is missing an entry point.

MCP (Model Context Protocol) is the open standard that closes that gap: it defines how an AI assistant discovers and calls tools exposed by an external system. An MCP Server is Fimaflow speaking that protocol — so instead of describing your workflow to an assistant, you let it look directly, within a scope you control.

A connection is read-only by default: it can list, inspect, and analyze. Check Allow this connection to run workflows when creating one to also let it trigger a real run — there's no way to change a workflow's nodes or edges through MCP yet. See Limitations.

How it works

Every workflow can generate its own MCP connection. Once connected, the assistant sees a fixed set of tools, scoped to that workflow's project if it has one. Most workflows don't — they only belong to a workspace — so most connections end up scoped to that whole workspace instead: the assistant can see every workflow in it, not just the one you generated the connection from. Either way, never your whole account, never other people's workspaces. See Permissions & safety.

Built-in tools, available on every connection:

• • about_fimaflow — what Fimaflow is and how to reason about a workflow. Worth calling first.
• • list_workflows — list the workflows in scope.
• • get_workflow — a workflow's full node and edge graph.
• • list_executions — a workflow's run history (status and timing per run).
• • get_execution_logs — the node-by-node trace of one execution, defaulting to the most recent — what each node did, or why one failed.
• • analyze_workflow — detected API routes, environment variables, and structural warnings (e.g. a flow with no entry point).
• • run_workflow — run a workflow for real and return the result. Only available if the connection was created with Allow this connection to run workflows checked — read-only otherwise.

Setting it up

Open a workflow in the builder, click MCP in the left sidebar. Check Allow this connection to run workflows if you want the assistant to be able to trigger a real run — leave it unchecked for a read-only connection — then click New connection. You get a single URL — shown once — add it to your assistant's MCP configuration exactly like you would for any other MCP server:

{
  "mcpServers": {
    "fimaflow": {
      "url": "https://mcp.fimaflow.com/<connection-secret>/<workflow-id>"
    }
  }
}

That's it — no separate token, no header to add. The URL itself is the credential; treat it like a password. Restart your assistant (or reload its MCP connections) and it should list the Fimaflow tools alongside whatever else it already has access to. See Integrations for the exact steps per assistant (Claude, Cursor, ChatGPT, JetBrains AI).

Connecting Claude Code

Claude Code can add the connection from the terminal — no config file to edit. Paste the URL you copied from the MCP panel, then check that it's connected:

❯ claude mcp add --scope user --transport http fimaflow "https://mcp.fimaflow.com/<connection-secret>/<workflow-id>"
❯ claude mcp list

claude mcp list should show fimaflow as connected. Start (or restart) a Claude Code session and ask it: "What Fimaflow tools do you have?"

• • --scope user makes the connection available in every project on your machine. Drop it to keep the connection to the current project only.
• • Don't use --scope project with this URL: it writes the server into a .mcp.json file meant to be committed, and the URL is a credential.
• • To remove it later: claude mcp remove fimaflow. Revoking the connection from the MCP panel also stops it working immediately.

Example: debugging a failed run

A teammate reports that Create Order failed overnight. Instead of opening the builder, they ask their assistant:

> "Why did the Create Order workflow fail last night?"

Assistant calls: list_workflows()
  → finds workflow "Create Order" (id: 7c1e...)

Assistant calls: get_execution_logs({ workflowId: "7c1e..." })
  → 3. code (validate-order)  ✗  Error: DATABASE_URL is not set

"The node failed because DATABASE_URL wasn't set — looks like the
 .ENV node's value was cleared. Want me to check the workflow's
 current config to confirm?"

The assistant only saw what the node already logs — never a credential or a raw connection string. See Debugging.

A second, lighter check — is a workflow even ready to export — uses analyze_workflow instead of waiting on a run:

> "Is the Create Order workflow ready to export as an API?"

Assistant calls: analyze_workflow({ workflowId: "7c1e..." })
  → warnings: ["No API Endpoint node found — the generated server would have no routes."]

"Not yet — the graph has no API Endpoint node, so exporting it now
 would produce a backend with zero routes."

Permissions & safety

• • A connection is scoped to the project of the workflow it was created from, or to its whole workspace if that workflow isn't filed into a project — the common case. Check which one applies before sharing a connection: workspace scope means every workflow there, not just the one you generated it from.
• • Either way, an assistant with a connection can't see your other projects, other workspaces, or anyone else's account.
• • Read-only unless you check Allow this connection to run workflows when creating it — an assistant can only trigger a real run if you explicitly granted that, per connection.
• • Node config comes back as stored — secret values (API keys, passwords) are never included, whether or not they're set. A run started by an assistant never exposes them either.
• • Revoke a connection at any time from the workflow's MCP panel; it stops working on the very next request.

Limitations

• • No way to edit a workflow's nodes or edges through MCP yet — an assistant can inspect and run a workflow, not change its graph. That's planned, not shipped.
• • No database query tool yet, and no way to expose a specific workflow as a named, callable tool of its own — also planned.
• • A connection can span every workflow in a workspace (see Permissions & safety above) — there's no way to scope one to a single workflow if that workflow isn't filed into a project.