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

# Connect MCP servers to iBlueprint Blueprint nodes

> Connect Model Context Protocol servers to iBlueprint Blueprints so AI nodes can query external tools, APIs, and data sources at runtime.

Model Context Protocol (MCP) is an open standard that lets AI models call external tools and retrieve structured context from services you define. In iBlueprint, you attach MCP servers directly to a Blueprint. When that Blueprint runs, its AI nodes can invoke any tool exposed by the server — querying databases, calling internal APIs, or fetching live data — without you having to hard-code those capabilities into the Blueprint itself.

## Add an MCP server to a Blueprint

<Steps>
  <Step title="Open Blueprint settings">
    Open the Blueprint you want to extend, then navigate to **Settings → MCP Servers** within the Blueprint editor.
  </Step>

  <Step title="Click Add server">
    Click **Add MCP server** to open the configuration panel.
  </Step>

  <Step title="Enter the server URL">
    Paste the full URL of your MCP server endpoint, for example:

    ```text theme={null}
    https://mcp.your-company.com/tools
    ```
  </Step>

  <Step title="Configure authentication">
    Choose an authentication type and supply the required credentials. See [Authentication types](#authentication-types) below.
  </Step>

  <Step title="Set a timeout">
    Enter the maximum time (in seconds) iBlueprint should wait for the server to respond. The allowed range is **1–120 seconds**; the default is **30 seconds**.
  </Step>

  <Step title="Add custom headers (optional)">
    If your server requires additional HTTP headers — for example, a workspace ID or tenant identifier — add them as key-value pairs in the **Custom headers** section.
  </Step>

  <Step title="Declare available tools (optional)">
    List the tool names your server exposes. This helps iBlueprint surface the right tools to AI nodes and avoids discovery requests at runtime. Each tool entry accepts a name, an optional description, and a JSON parameters schema.
  </Step>

  <Step title="Save">
    Click **Save**. The server is now available to all AI nodes in this Blueprint.
  </Step>
</Steps>

## Authentication types

iBlueprint supports five authentication schemes for MCP servers.

<Tabs>
  <Tab title="None">
    No authentication. Use this for internal servers that are already protected at the network level.

    No additional fields are required.
  </Tab>

  <Tab title="Bearer token">
    Sends an `Authorization: Bearer <token>` header with every request.

    **Required fields:**

    * **Token** — your bearer token string.

    ```text theme={null}
    Authorization: Bearer eyJhbGciOiJSUzI1NiJ9...
    ```
  </Tab>

  <Tab title="API key">
    Sends your API key in a custom header.

    **Required fields:**

    * **API key** — the key value.
    * **Header name** — the header your server expects (for example, `X-API-Key`).

    ```text theme={null}
    X-API-Key: sk-live-abc123
    ```
  </Tab>

  <Tab title="Basic auth">
    Encodes a username and password using HTTP Basic authentication.

    **Required fields:**

    * **Username**
    * **Password**

    ```text theme={null}
    Authorization: Basic dXNlcjpwYXNz
    ```
  </Tab>

  <Tab title="OAuth">
    Delegates authentication to an OAuth provider.

    **Required fields:**

    * **Provider** — the name of your OAuth provider.
    * **Account ID** — the account or tenant identifier used during the OAuth flow.
    * **Require auth** — toggle on to enforce that a valid OAuth token is present before any tool call is made.
  </Tab>
</Tabs>

## Timeout configuration

The timeout controls how long iBlueprint waits for a single response from the MCP server before failing the tool call.

| Setting | Value       |
| ------- | ----------- |
| Minimum | 1 second    |
| Maximum | 120 seconds |
| Default | 30 seconds  |

Choose a lower timeout for fast, latency-sensitive tools. Increase it for servers that perform expensive computations or call slow downstream APIs.

<Warning>
  A timed-out tool call causes the AI node to receive an error rather than a result. Design your MCP server to return partial results quickly, or break large operations into smaller sequential tool calls.
</Warning>

## Manage MCP servers

You can enable, disable, or delete any MCP server from the Blueprint's **Settings → MCP Servers** list without modifying the Blueprint graph itself. Disabling a server keeps its configuration intact but prevents AI nodes from calling it.

<Tip>
  Use the **tools list** field to pre-declare the tools your server provides. This avoids a discovery round-trip at the start of each Blueprint run and makes tool selection more predictable inside AI nodes.
</Tip>