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

# MCP mode

> Run AgentCash as an MCP server inside any compatible AI client.

AgentCash can run as a [Model Context Protocol](https://modelcontextprotocol.io) (MCP) server, giving your AI client direct tool access to all AgentCash capabilities: paid API calls, endpoint discovery, wallet management, and authenticated requests.

MCP mode is one of three ways to use AgentCash, alongside the [CLI](/cli/overview) and the [agent skill](/quickstart-agents#2-give-your-agent-the-skill).

## When to use MCP mode

MCP mode is ideal when your AI client supports MCP and you want:

* **Tool-based access**: the agent calls AgentCash tools directly
* **Structured responses**: tool responses come back as structured data the agent can parse
* **Automatic integration**: the tools are available in the agent's context with no extra setup

## Quick install

The guided installer detects your client:

```bash theme={null}
npx agentcash install
```

Or install for a specific client:

<Tabs>
  <Tab title="Claude Code">
    ```bash theme={null}
    claude mcp add agentcash --scope user -- npx -y agentcash@latest
    ```
  </Tab>

  <Tab title="Cursor">
    ```bash theme={null}
    npx agentcash install --client cursor
    ```

    Or add manually to `.cursor/mcp.json`:

    ```json theme={null}
    {
      "mcpServers": {
        "agentcash": {
          "command": "npx",
          "args": ["-y", "agentcash@latest"]
        }
      }
    }
    ```
  </Tab>

  <Tab title="Claude Cowork">
    Follow the [guided install](https://agentcash.dev/install/claude) to download and install the MCPB bundle.
  </Tab>

  <Tab title="Claude Desktop">
    Follow the [guided install](https://agentcash.dev/install/claude), or add manually to your config file (`~/Library/Application Support/Claude/claude_desktop_config.json` on macOS):

    ```json theme={null}
    {
      "mcpServers": {
        "agentcash": {
          "command": "npx",
          "args": ["-y", "agentcash@latest"]
        }
      }
    }
    ```

    Restart Claude Desktop after saving.
  </Tab>

  <Tab title="Other clients">
    See [full client list](/installation/other-clients) for Codex, Windsurf, Warp, and more.
  </Tab>
</Tabs>

## Available tools

When running in MCP mode, AgentCash exposes these tools to your agent:

| Tool                                                      | What it does                                       |
| --------------------------------------------------------- | -------------------------------------------------- |
| [`fetch`](/tools/fetch)                                   | HTTP request with automatic SIWX auth and payment  |
| [`fetch_with_auth`](/tools/fetch-with-auth)               | Deprecated alias for `fetch`                       |
| [`get_balance`](/tools/get-balance)                       | Total wallet balance across supported networks     |
| [`list_accounts`](/tools/list-accounts)                   | Per-network balances, addresses, and deposit links |
| [`search`](/tools/search)                                 | Find API providers by natural-language query       |
| [`check_endpoint_schema`](/tools/check-endpoint-schema)   | Probe pricing and schema without paying            |
| [`discover_api_endpoints`](/tools/discover-api-endpoints) | Find all endpoints on an origin                    |
| [`redeem_invite`](/tools/redeem-invite)                   | Redeem an invite code for free USDC                |

## Verify installation

Start a session with your AI client and ask:

> What is my AgentCash balance?

The agent should call the `get_balance` tool and return your total balance. If you later need deposit links or network-specific addresses, the agent should call `list_accounts`.

## Server options

You can pass flags to the MCP server process:

| Flag               | Description                           |
| ------------------ | ------------------------------------- |
| `--verbose` / `-v` | Enable debug output on stderr         |
| `--quiet` / `-q`   | Suppress all stderr output            |
| `--dev`            | Use development endpoints (localhost) |
