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

# Connect via MCP

> Connect the Concierca AI MCP server from Claude, ChatGPT, or any MCP client — endpoint, authentication, and client setup.

Concierca AI exposes its whole surface — agents, missions, data, knowledge, memory, skills, approvals, and the market-intelligence vertical — as a **Model Context Protocol (MCP)** server. Any MCP-capable client (Claude, ChatGPT, your own agent) connects once and can then read and act inside your workspace, always under the same tenant, scope, and governance rules as the app.

<Note>
  **Building an AI client?** After connecting, read [the Operating Contract](/api/ai/operating-contract) — the session-start guide for how an AI should ground itself and act inside Concierca.
</Note>

<Note>
  Every call runs **server-side as the user who owns the key**. Reads are filtered to what that user may see; writes need the matching per-tool grant. Secrets and OAuth flows never travel the model channel.
</Note>

## Endpoint

| Surface                                                | URL                                            |
| ------------------------------------------------------ | ---------------------------------------------- |
| **Main MCP** (all tools, incl. the federated vertical) | `https://app.concierca.ch/api/mcp/v1`          |
| **Vertical only** (Watch Market Intelligence)          | `https://app.concierca.ch/api/mcp/watchinside` |

The main endpoint is what you connect in day-to-day use. The vertical endpoint is a standalone surface for the market-intelligence tools (`vertical_*`) — see [Vertical](/api/vertical).

## Authentication

Two ways to connect, both from **Settings → MCP Access**:

1. **OAuth (recommended for a person)** — connect the client and approve in the browser. A Company-Admin connect grants the full surface; a Member connect grants the read-only baseline. See [Authentication](/api/authentication) for the scope model.
2. **API key** — generate a key and pass it as a bearer token. Best for headless agents and the [CLI](/api/cli).

## Connect from Claude (Desktop / Code)

Add the server to your MCP client config, pointing at the main endpoint and your key:

```json theme={null}
{
  "mcpServers": {
    "concierca": {
      "url": "https://app.concierca.ch/api/mcp/v1",
      "headers": { "Authorization": "Bearer <your-mcp-key>" }
    }
  }
}
```

For OAuth-capable clients, add the server by URL and complete the browser approval instead of pasting a key.

## Connect from ChatGPT

Add a **connector** pointing at `https://app.concierca.ch/api/mcp/v1` and authorize it.

<Warning>
  ChatGPT loads the tool list only when a connector is **created**, and does not refresh it live. After your grants change (new tools, a reconnect), **delete the connector and add it again** to pick up the current tool set — re-login or "reload" is not enough.
</Warning>

## First calls

Ground every session before acting:

1. `whoami` — confirms identity, role, tenant, and the tools your key may use.
2. `get_context` — returns your Operating Manual (the always-load governance + knowledge block).
3. `guardrails_get` — the governance rules to honor; over MCP these are pull-only, so fetch them at session start.

Then use the resource tools — [All Tools](/api/tools) is the full 136-tool index; [Data](/api/data), [Agents](/api/agents), [Missions](/api/missions), [Knowledge](/api/knowledge), [Memory](/api/memory), [Skills](/api/skills), [Inbox](/api/inbox), [Approvals](/api/approvals), [Administration](/api/admin), [Vertical](/api/vertical). Naming, common parameters, errors, and limits are in [Conventions](/api/conventions).

If you're driving Concierca as an AI client, the [For AI Agents](/api/ai/operating-contract) guide turns these first calls into a full operating contract.

## Reconnecting

Reconnecting inherits your existing scopes and deactivates the previous key — you never lose access by reconnecting, and old keys stop working. Details in [Authentication](/api/authentication).