Postman MCP Server

Postman's official MCP server connects an AI coding assistant to your Postman account: workspaces, collections, API specifications and environments become things the agent can read and change directly.

Four configurations, not one

The tool count is the usual failure mode for a large API, and Postman addresses it by shipping tiers rather than one 100-tool server:

• Minimal (default) — the essential operations for working on a single collection, workspace or environment.
• Code — adds tooling to generate well-organised client code from public and internal API definitions. This is the tier to use when the goal is consuming an API rather than curating it.
• Full — all 100+ Postman API tools, including the Enterprise collaboration surface.
• Learn — searches the Postman documentation for guides and reference.

Picking the narrowest tier that covers the job keeps the agent's tool list short and its choices better.

Remote and local

The remote server at https://mcp.postman.com supports OAuth compliant with the MCP authorization specification, which means no API key handling at all — append /minimal, /code or /mcp to select the tier. The local server runs as npx @postman/postman-mcp-server with POSTMAN_API_KEY set, plus --code or --full to change tier. Local is the one you need for testing an API running on your own machine, since the remote server cannot reach your workstation.

EU customers get https://mcp.eu.postman.com/… for the remote server or --region eu locally; the EU endpoints are API-key only, with no OAuth.

What people use it for

Running a collection against an API continuously while developing it, keeping code and collections in sync, bulk collection and documentation edits, creating specs from existing code and generating collections from them, and producing client code that follows the project's conventions.

Officially supported across Claude Code, Claude Desktop, Cursor, VS Code, Codex, GitHub Copilot CLI, Kiro, Antigravity CLI and Docker. Apache-2.0 licensed.