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

# Connect your AI client to Plainrouter

> Find Plainrouter setup for Claude, ChatGPT, Cursor, Grok, Gemini CLI, Meta Muse, Hermes, OpenClaw, n8n and Make. Start with synthetic data.

Choose your client to try Plainrouter’s account-free sandbox, then connect your workspace. Chat clients, terminal agents and workflow platforms use different configuration screens; the server and workspace boundaries stay the same.

## Choose your setup guide

| Client | Setup path |
| - | - |
| [Claude](/docs/mcp/setup#start-in-test-mode) | Claude Code terminal configuration and first read. |
| [ChatGPT](/docs/mcp/overview#connect-chatgpt) | Custom MCP app with OAuth for production. |
| [Cursor](/docs/mcp/clients/cursor) | Cursor Agent and user-level MCP configuration. |
| [Grok](/docs/mcp/clients/grok) | Custom connector in Grok. |
| [Gemini](/docs/mcp/clients/gemini) | Gemini CLI; not the Gemini web app. |
| [Meta Muse](/docs/mcp/clients/muse) | Muse web custom connector; prompt-led setup, not Muse Code. |
| [Hermes](/docs/mcp/clients/hermes) | Nous Research Hermes Agent. |
| [OpenClaw](/docs/mcp/clients/openclaw) | The OpenClaw installation or Gateway running your agent. |
| [n8n](/docs/mcp/clients/n8n) | MCP Client Tool attached to an AI Agent; separate HTTP conversion workflow. |
| [Make](/docs/mcp/clients/make) | MCP Client scenario modules; separate HTTP conversion workflow. |

## Endpoints and access

| Connection | Endpoint | Authentication |
| - | - | - |
| Sandbox | `https://plainrouter.com/mcp/sandbox` | None. Synthetic data, no Plainrouter account, never contacts Meta. |
| Production | `https://plainrouter.com/mcp` | Streamable HTTP with `Authorization: Bearer …` using a Workspace key, or client-supported Plainrouter OAuth sign-in. |

[Create a Workspace key](/docs/mcp/workspace-tokens) for one workspace. Read covers analysis; Actions require Write, including their record-read tools. Keep credentials in protected client settings, never prompts or source control. OAuth clients obtain workspace-scoped execution credentials through sign-in instead of requiring a pasted key.

**Ask** waits for approval; **Full** runs policy-allowed changes automatically. Rule violations are blocked either way. Client permission prompts do not override [Plainrouter policy](/docs/actions/policies-and-safety).

## What proves the connection worked?

Inspect a real tool result. In the sandbox, a successful read contains `sandbox: true`. In production, a read must return the intended workspace. A discovered tool list or an assistant saying “connected” does not prove authenticated access. Agent wording can vary, and connecting a client does not verify conversion delivery or ad execution.

For another client, use the [transport and authentication reference](/docs/mcp/setup#which-mcp-protocol-should-my-client-use). For conversion sending without an agent, start with [n8n and Make automation](https://plainrouter.com/solutions/automation).


This documentation is built and hosted on [Mintlify](https://mintlify.com), a developer documentation platform.