MCP on OpenAI Agent Builder

OpenAI Agent Builder is a visual canvas for composing agent workflows. Use the MCP tool node to call Onboard tools (search customers, list projects, post comments, and more) from your agents.

Before you start

Complete the requirements on MCP setup (Admin access, MCP plan, API key from Integrations → API).

Add the Onboard MCP server

When you add a tool to a workflow:

1. Drag or insert an MCP node from the Tools palette (or add MCP to an agent’s tools).
2. Click MCP → + Add → + Server (under third-party servers).
3. In Connect to MCP Server, enter:

4. Click Connect. Agent Builder should list Onboard tools after the server connects.

Production example with Onboard’s hosted SSE URL and API key auth:

Use MCP in a workflow

After the server is connected, configure the MCP node:

• Tool — pick the Onboard tool to invoke (for example search_customers, list_active_projects, or post_comment).
• Require approval — set to Always for write actions or sensitive reads until you trust the workflow in production.

Intent-first workflow tools are usually easier for agents than raw [API] tools. Start with tools that match your use case; use browse_full_api and run_api_request only when you need an endpoint not wrapped as a workflow tool.

Example workflow with an MCP node wired to Onboard:

See MCP setup for the full endpoint list.

OpenAI documentation

Agent Builder’s MCP node and connector options are described in OpenAI’s Node reference — MCP and MCP and Connectors guides.

Related

• MCP setup — base URL, /mcp paths, authentication
• Connecting to Onboard MCP — API key vs OAuth
• Security checklist
• MCP desktop clients