Coda has an assistant built in, but you can also use your own AI client. Point Claude, ChatGPT or any other MCP client at the URL below, describe the pipeline you want, and you get back a link that opens the workflow in Coda.
MCP (the Model Context Protocol) lets an AI client use external tools.
It reads Coda’s node catalogue, adds the nodes, connects them and sets their parameters. It then checks the workflow using the same checks Coda runs on the canvas, and fixes any problems before it answers.
The link opens the workflow in Coda in your browser. From there you can run it (with your own credentials) and edit it like any other workflow. The server only builds the workflow, it never runs it.
The server reads the node definitions from the deployed version of Coda, so new nodes are available to the model as soon as they go live. There is nothing to upgrade.
^LC4$ →
Connectivity, downstream → Group By postType → Bar Chart. Open it and press Run:
The server is hosted, so there is nothing to install and no key to paste. Just add this URL wherever your client keeps its remote MCP servers:
One line in a terminal:
claude mcp add --transport http coda https://flyem.mrc-lmb.cam.ac.uk/coda-mcp/mcp
Any MCP client can add a remote server by URL, either in its connector settings or in its config file (see below).
Then just ask. The model first calls coda_guide, which returns the full catalogue of node types with their ports and parameters, so you don’t need to explain Coda to it.
Add the URL to your client, then ask it for a connectome workflow. If you would rather stay inside Coda, the built-in assistant does the same job using your own API key.