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

# Tools

> Add MCP server and web search capabilities to agents

> Integrate external tools into your agents via Model Context Protocol (MCP) servers or web search.

<Tabs>
  <Tab title="MCP Server">
    ## `agno/tools/mcp`

    Connects to a Model Context Protocol (MCP) server, giving agents access to external tools.

    ### Config

    | Field                         | Type                                        | Required    | Default       | Description                                                                          |
    | ----------------------------- | ------------------------------------------- | ----------- | ------------- | ------------------------------------------------------------------------------------ |
    | `command`                     | `string`                                    | Conditional | —             | Command to run the MCP server (for `stdio` transport). Mutually exclusive with `url` |
    | `url`                         | `string`                                    | Conditional | —             | URL for SSE or streamable-http transport. Mutually exclusive with `command`          |
    | `transport`                   | `"stdio"` \| `"sse"` \| `"streamable-http"` | No          | Auto-detected | Transport protocol. Inferred from `command` (stdio) or `url` (sse) if not set        |
    | `env`                         | `dict[string, Field[string]]`               | No          | —             | Environment variables for the server process                                         |
    | `headers`                     | `dict[string, Field[string]]`               | No          | —             | Static HTTP headers for remote transports                                            |
    | `include_run_context_headers` | `boolean`                                   | No          | `false`       | Forward agent run context as headers (`X-User-ID`, `X-Session-ID`, `X-Run-ID`)       |
    | `timeout_seconds`             | `integer`                                   | No          | `10`          | Connection timeout in seconds                                                        |
    | `include_tools`               | `list[string]`                              | No          | —             | Only include these tool names                                                        |
    | `exclude_tools`               | `list[string]`                              | No          | —             | Exclude these tool names                                                             |
    | `tool_name_prefix`            | `string`                                    | No          | —             | Prefix added to all tool names (avoids collisions)                                   |

    ### Outputs

    | Field  | Type     | Description                                          |
    | ------ | -------- | ---------------------------------------------------- |
    | `spec` | `object` | Serialized MCP tools spec for runtime reconstruction |

    ### Example — Remote MCP server

    ```yaml theme={"theme":{"light":"min-light","dark":"min-dark"}}
    provider: agno
    resource: tools/mcp
    name: my-tools
    config:
      url: "https://mcp.example.com/sse"
      headers:
        Authorization:
          provider: pragma
          resource: secret
          name: mcp-token
          field: outputs.TOKEN
      include_run_context_headers: true
      timeout_seconds: 30
    ```

    ### Example — Local stdio server

    ```yaml theme={"theme":{"light":"min-light","dark":"min-dark"}}
    provider: agno
    resource: tools/mcp
    name: local-tools
    config:
      command: "npx -y @modelcontextprotocol/server-filesystem /data"
      env:
        API_KEY:
          provider: pragma
          resource: secret
          name: api-key
          field: outputs.API_KEY
    ```
  </Tab>

  <Tab title="Web Search">
    ## `agno/tools/websearch`

    Adds web search capabilities to agents.

    ### Config

    | Field               | Type      | Required | Default  | Description                                                                        |
    | ------------------- | --------- | -------- | -------- | ---------------------------------------------------------------------------------- |
    | `enable_search`     | `boolean` | No       | `true`   | Enable web search                                                                  |
    | `enable_news`       | `boolean` | No       | `true`   | Enable news search                                                                 |
    | `backend`           | `string`  | No       | `"auto"` | Search backend: `auto`, `duckduckgo`, `google`, `bing`, `brave`, `yandex`, `yahoo` |
    | `modifier`          | `string`  | No       | —        | Text prepended to all queries (e.g., `"site:example.com"`)                         |
    | `fixed_max_results` | `integer` | No       | —        | Override default max results                                                       |
    | `proxy`             | `string`  | No       | —        | Proxy URL for requests                                                             |
    | `timeout`           | `integer` | No       | `10`     | Request timeout in seconds                                                         |
    | `verify_ssl`        | `boolean` | No       | `true`   | Verify SSL certificates                                                            |

    ### Outputs

    | Field              | Type           | Description                                           |
    | ------------------ | -------------- | ----------------------------------------------------- |
    | `pip_dependencies` | `list[string]` | Required Python packages (`ddgs>=8.0.0`)              |
    | `spec`             | `object`       | Serialized web search spec for runtime reconstruction |

    ### Example

    ```yaml theme={"theme":{"light":"min-light","dark":"min-dark"}}
    provider: agno
    resource: tools/websearch
    name: search
    config:
      enable_search: true
      enable_news: true
      backend: duckduckgo
    ```
  </Tab>
</Tabs>

## Dependencies

**Depends on:** Nothing directly, but `env` and `headers` fields on MCP tools support FieldReferences.

**Depended on by:**

* `agno/agent` — in the `tools` list
* `agno/team` — in the `tools` list

## Notes

* For MCP tools, you must provide exactly one of `command` or `url`.
* The `transport` type is auto-detected if not specified: `command` implies `stdio`, `url` implies `sse`.
* The `include_run_context_headers` option forwards the agent's user ID, session ID, and run ID as HTTP headers to the MCP server.
* Web search requires the `ddgs` package, which is automatically included as a pip dependency.