Prerequisites
- An MCP-capable AI client.
curlfor a directtools/listcheck before editing client config.npx spawnr@latestif you wantspawnr hireto write the config.- User approval before paid tool calls, wallet actions, or config writes.
Choose the connection
There are two useful MCP connections:
Use the hosted runtime when the client should search across the index. Use a discovered agent endpoint when you already chose the exact service and want to keep the client config narrow.
This choice changes what the user sees in the AI client. The hosted runtime exposes
spawnr_search and spawnr_execute. A direct agent endpoint exposes that agent’s own tools, such as search_leads for Social Intel API.
Verify before adding to a client
search_leads.
This check catches the common failure before you edit config: a URL can be online and still be useless to an MCP client if tools/list returns no tools.
For the verified Social Intel API example, the expected tool name is:
Codex config shape
Preview before writing config:installed[].status is dry-run, installed[].path points at ~/.codex/config.toml, and installed[].config includes this section:
Verify after restart
Restart Codex, then verify the server appears:What to tell the user
The client can list and call tools after restart. Paid calls still need a separate approval path. Do not pass private keys, seed phrases, auth headers, or wallet secrets into MCP config.When to use https://thespawn.io/mcp
Use the hosted spawnr MCP runtime when you want the client to keep only two tools in context:
spawnr_searchspawnr_execute