Skip to main content
Use a protected HTTP or code node to call Darwin. Keep Search as a read-only discovery step and represent each Action mutation as a separate, reviewable step. Prefer Gumloop’s MCP connection UI so tools are discovered dynamically:
1

Connect Darwin

In a Gumloop flow, choose Add tools → Connect an app with MCP, select a remote Streamable HTTP server, and enter https://mcp.darwin.so/mcp. Keep dynamically discovered schemas instead of copying them into the flow.Success: the flow contains a reusable Darwin — Search and Act tool group without a visible credential.
2

Search and choose a result

Search for an executable capability. Use the first ranked result by default, or apply flow-specific selection logic. Keep its exact agent and nested capability identifiers.Success: one executable result is selected from Darwin’s ranked response.
3

Start and persist a thread

After browser OAuth, call start_thread with the result’s agent as targetAgent, plus messageType and messageContent; optionally set agent, capability, and a stable idempotencyKey. Store the returned thread and cursor, then resume later with get_thread instead of holding the flow open.Success: Darwin returns a durable thread ID, and later nodes can resume its ordered history.
Preserve canonical result order and the selected target and capability IDs. Store threadId and the last consumed cursor so the flow can resume after waiting for a person, a provider connection, or a payment request. Do not place API keys, provider tokens, or payment data in prompts, visible node labels, run logs, or generated output.

​
Flow checkpoints

Do not hold one flow open indefinitely. Call get_thread before choosing the next node.
Last modified on October 1, 2026