MCP Client

Why this exists

MCP isn't specific to AI assistants — it's just a standard way to expose a set of tools over a connection. Once another team, or a third-party product, exposes its own capabilities as an MCP server, your workflow can call them the same way an assistant would: no bespoke integration, no separate API client to maintain.

Use it instead of an API node whenever the thing you're calling already speaks MCP — you get a typed, self-describing set of tools instead of a hand-rolled HTTP request.

How it works

Drop an MCP Client node onto the canvas and point it at the target server's URL and credentials — usually stored in an Environment Variable so they never sit in plain text on the node. On first connection, the node fetches the server's list of tools; pick one and map data from earlier in the workflow onto its inputs, exactly like configuring any other node.

• • mcpServer — connection details of the external server.
• • tool — which of the server's exposed tools to call.
• • arguments — the tool's input, built from data flowing through the workflow.

The tool's response becomes the node's output, ready to transform, store, or return like any other node's result.

Example: enriching an order with inventory data

Say your warehouse system exposes its own MCP server with a check_stock tool. Extend the Create Order workflow to check stock before confirming:

API Endpoint  (POST /orders)
   ↓
Validation
   ↓
MCP Client    tool: check_stock
              arguments: { sku: "{{sku}}", qty: "{{qty}}" }
   ↓
Conditional   available === true ?
   ↓                              ↓
  SQL (insert)              Response (409 out of stock)

The warehouse team never had to build a REST endpoint just for this — their existing MCP server already covered it.

When to reach for it

• • Call tools already exposed by another internal team's MCP server, instead of asking them to build a REST endpoint for you.
• • Integrate a third-party product that ships an official MCP server, without writing a custom client.
• • Chain a Fimaflow workflow into a larger multi-agent system where every step is itself MCP-callable.

Limitations

• • The node calls one tool per step — chaining several tool calls from the same server means several MCP Client nodes in sequence.
• • If the external server is unreachable or the tool errors, the node fails like any other — see Debugging to inspect what it received back. Pair it with a Try/Catch node if the external server is expected to be flaky.
• • The server's tool list is fetched once when you add the node — refresh it manually if the external server adds new tools later.