> Toden documentation. For the full index fetch /llms.txt. Every page is available as Markdown by appending .md to its URL. Identity verification uses /api/kyc, Risk Intelligence uses /api/v1; they are separate contracts with different error shapes. Never invent an endpoint, field or enum value: read the API reference page for the operation first.

# MCP server

> How to connect an AI client to Toden's documentation MCP server, what it can and cannot do, and the status of an MCP server that works with your account.

The Model Context Protocol (MCP) lets an AI client such as Claude or Cursor connect to a server and call its tools. This documentation site serves one, so your agent can look things up in the Toden docs without you pasting pages into the chat.

|                                                        | Status                                   |
| ------------------------------------------------------ | ---------------------------------------- |
| **Documentation MCP server** (search these docs)       | Available at `/_mcp/server` on this site |
| **Account MCP server** (create sessions, read results) | Planned. Not available yet.              |

## What the documentation server does

It has one tool, `searchDocs`: the agent sends a question and gets back the matching passages with links to their pages.

| It does                                                   | It does not                                    |
| --------------------------------------------------------- | ---------------------------------------------- |
| Search the guides and the API reference                   | Read your account, customers or verifications  |
| Return passages and page links for the agent to work from | Create sessions, run checks or change settings |
| Stay current with the published docs                      | Take or need a Toden API key                   |

It gives an agent the same public pages you can read here, and nothing else.

## Connect your client

The server address is this site's address followed by `/_mcp/server`.

#### Claude Code

```bash
claude mcp add --transport http toden-docs https://docs.toden.co/_mcp/server
```

Then run `/mcp` inside Claude Code to check the connection.

#### Cursor

Add the server to `.cursor/mcp.json` in your project, or to your global Cursor MCP settings:

```json
{
  "mcpServers": {
    "toden-docs": {
      "url": "https://docs.toden.co/_mcp/server"
    }
  }
}
```

#### Other clients

Any client that supports remote MCP servers over HTTP takes the same address. Follow your client's own instructions for adding a remote server.

Once connected, ask in plain language, for example "How do I verify a Toden webhook signature?", and the agent looks the answer up in these docs.

> **Tip**
>
> For building an integration, pair the server with the [integration prompt](/get-started/ai/integration-prompt). The prompt tells the agent what to ask you and which rules to follow; the server lets it check each detail against the docs.

## Without MCP

Everything the server searches is also available in forms an agent can read directly.

#### [llms.txt and Markdown pages](/get-started/ai/machine-readable)

An index of every page, and each page as Markdown by adding `.md` to its address.

#### [OpenAPI contracts](/get-started/ai/machine-readable)

The exact request and response shapes for both API families.

#### [Integration prompt](/get-started/ai/integration-prompt)

A prompt to paste into your agent, with the rules that are easy to get wrong.

#### [Agent skills](/get-started/ai/overview#agent-skills)

One short brief per product to save in your agent's skills folder.

## Is there an MCP server I can run myself?

Not as a product. Toden's own engineers use a small local MCP server while working on the platform. It runs on a developer's machine from the Toden source code, serves integration documentation only, and makes no API calls. It is not published as a package and is not available to customers.

You do not need it. The same material is public through `llms.txt`, the Markdown pages and the OpenAPI contracts above.

## An MCP server that works with your account

An MCP server that performs actions, such as creating a verification session or reading a result, is planned and not available yet. We will announce it here and in the [changelog](/changelog).

To let an agent work with the API today, have it write ordinary code against the REST API using the OpenAPI contracts, and run that code in Sandbox with a test key. Keep these rules:

* Never paste a secret key (`sk_...`) into a prompt or a chat. Read keys from environment variables.
* Build and test in Sandbox. A person should approve anything that affects real customers.
* A failed or timed-out call is never an approval.

The full list is in [Integrate with an AI agent](/get-started/ai/overview).

## Related

* [Integrate with an AI agent](/get-started/ai/overview)
* [Machine-readable docs](/get-started/ai/machine-readable)
* [Integration prompt](/get-started/ai/integration-prompt)