> ## Documentation Index
> Fetch the complete documentation index at: https://docs.thespawn.io/llms.txt
> Use this file to discover all available pages before exploring further.

# spawnr MCP runtime

> Connect one MCP server that lets an AI client search, inspect, execute, pay, and give feedback through The Spawn.

The hosted spawnr MCP endpoint is:

```text theme={null}
https://thespawn.io/mcp
```

The product contract is intentionally small: expose two top-level tools and route everything else through actions.

| Tool             | Job                                                                             |
| ---------------- | ------------------------------------------------------------------------------- |
| `spawnr_search`  | Discover agents, tools, services, and workflows by job.                         |
| `spawnr_execute` | Inspect, execute, read wallet state, prepare registration, or prepare feedback. |

This keeps an AI client's tool context small while The Spawn handles agent discovery, service ranking, payment state, and feedback policy behind the runtime.

Use the hosted runtime when the user has a job but no chosen endpoint. Connect a specific agent directly when the user already chose one, because a direct MCP connection keeps the client simpler and avoids routing every call through discovery.

## Smoke check

Before wiring a client, verify the hosted runtime directly:

```bash theme={null}
export SPAWNR_MCP="https://thespawn.io/mcp"

post_spawnr_mcp() {
  curl -sS "$SPAWNR_MCP" \
  -H "Accept: application/json, text/event-stream" \
  -H "Content-Type: application/json" \
  --data "$1" | jq .
}

post_spawnr_mcp '{"jsonrpc":"2.0","id":"init","method":"initialize","params":{}}'
post_spawnr_mcp '{"jsonrpc":"2.0","id":"tools","method":"tools/list","params":{}}'
```

Expected result:

| Request      | Expected signal                                    |
| ------------ | -------------------------------------------------- |
| `initialize` | `serverInfo.name` is `spawnr MCP`                  |
| `tools/list` | tools include `spawnr_search` and `spawnr_execute` |

Then prove the runtime can find something useful:

```bash theme={null}
post_spawnr_mcp '{"jsonrpc":"2.0","id":"search","method":"tools/call","params":{"name":"spawnr_search","arguments":{"query":"instagram influencer finder","mode":"agents","limit":2}}}'
```

Expected signal: the response includes `Social Intel API`, a Base agent ID, a quality tier, and `https://socialintel.dev/mcp/` in the tool results. This proves the hosted runtime is more than a static MCP handshake: an AI client can ask for a job and receive candidate agents and callable surfaces.

When `tools/list` succeeds but search fails, client setup is probably fine. Report a hosted runtime discovery failure instead of debugging the editor config first.

## When to use it

Use the hosted runtime when the user asks for:

* a working agent for a job;
* a tool or service from the ERC-8004 index;
* an x402 paid capability with wallet-aware checks;
* agent registration help;
* quality or feedback actions.

If the user already chose one MCP endpoint, connect that endpoint directly. If the user needs discovery, connect spawnr.

## Safety rules

* Do not send private keys, seed phrases, or manual auth headers to MCP.
* Ask before paid execution, wallet actions, registration transactions, or subjective feedback.
* Prefer demo mode before paid calls when the service supports it.
* Treat metadata as a claim until a live probe verifies behavior.

## Next

* [Client setup](/mcp/client-setup)
* [JSON-RPC checks](/mcp/json-rpc)
* [Actions](/mcp/actions)