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

# Latch MCP

> Use the Latch MCP to access the Latch platform from AI tools

## Overview

The [Model Context Protocol (MCP)](https://modelcontextprotocol.io/docs/getting-started/intro) is an open standard for connecting AI applications to external systems. You can use Latch's remote MCP server to securely connect your favorite AI development tools to the Latch platform. You can then directly access Latch resources and carry out actions on the Latch platform.

Using the Latch MCP requires you to have a Latch account. While there is no separate pricing for the MCP, you will be charged normally for any actions you perform on Latch using the MCP. For example, if you use the MCP to launch a [Latch workflow](https://wiki.latch.bio/workflows/overview), you will be charged the same as if you had launched it directly from the Latch Console or SDK.

## Setup

Authenticating to the Latch MCP only authorizes the connected AI client to access the MCP server. These credentials cannot be used for general Latch access through the Latch SDK or console.

<Tabs>
  <Tab title="Claude">
    You can install the Latch MCP as a third-party connector in Claude (web) or Claude Desktop. Follow the steps in [Claude's documentation to set up a third-party connector](https://claude.com/docs/connectors/custom/remote-mcp). When asked to provide the remote MCP URL, use `https://mcp.latch.bio/mcp`. Once you have set up the connector, you will be asked to authorize the connector to access your Latch account.

    <Frame>
      <img src="https://mintcdn.com/latchbio/xvrkgKJgLBYyVrF3/images/claude_mcp_auth.png?fit=max&auto=format&n=xvrkgKJgLBYyVrF3&q=85&s=cdd504040cf461e91d79111016b1c28b" alt="Claude connector setup for Latch MCP" width="463" height="480" data-path="images/claude_mcp_auth.png" />
    </Frame>
  </Tab>

  <Tab title="Claude Code">
    Run this command in your terminal:

    ```bash theme={null}
    claude mcp add --transport http latchbio https://mcp.latch.bio/mcp
    ```

    Then authenticate by running `/mcp` in Claude Code and following the OAuth flow.
  </Tab>

  <Tab title="Codex">
    First add the Latch MCP to Codex from your terminal:

    ```bash theme={null}
    codex mcp add latchbio --url https://mcp.latch.bio/mcp
    ```

    Then log in to the Latch MCP server:

    ```bash theme={null}
    codex mcp login latchbio
    ```
  </Tab>

  <Tab title="Cursor">
    Add this to your `~/.cursor/mcp.json` file:

    ```json theme={null}
    {
      "mcpServers": {
        "LatchBio": {
          "url": "https://mcp.latch.bio/mcp"
        }
      }
    }
    ```

    In Cursor, navigate to **Cursor Settings > Tools & MCP(s)**. You should see the `LatchBio` MCP server listed under `MCP Servers`. Click **connect** and proceed with the OAuth flow.
  </Tab>

  <Tab title="Other">
    Follow the MCP setup instructions for your application using `https://mcp.latch.bio/mcp` as your remote MCP server url.
  </Tab>
</Tabs>

## Tools

The Latch MCP server exposes tools that allow an AI agent to interact with the Latch platform. By default, Claude will ask you for permission before launching workflow executions.

| Tool                  | Description                                                                                                                |
| --------------------- | -------------------------------------------------------------------------------------------------------------------------- |
| `list_files`          | Lists the immediate contents of a directory in Latch Data. File contents are not returned.                                 |
| `get_file`            | Returns access information for a file stored in Latch Data, either as a Latch Console link or as a presigned download URL. |
| `list_workspaces`     | Lists the workspaces the current user can access, including the default workspace.                                         |
| `list_workflows`      | Discovers workspace workflows and public workflows that can be launched.                                                   |
| `get_workflow_schema` | Fetches launch metadata and parameter schema for a workflow.                                                               |
| `launch_workflow`     | Launches a workflow with parameter values matching the workflow schema.                                                    |
| `list_executions`     | Lists workflow executions in a workspace, with optional filters for name, workflow, and status.                            |
| `get_execution`       | Fetches workflow execution status, task nodes, result files, and a Console URL.                                            |
| `get_task_logs`       | Fetches inline task logs or a presigned download URL for full task logs.                                                   |