Summary
Import a Postman v2.1 collection and every request becomes an MCP tool for Claude, ChatGPT and Copilot: folders, auth and variables carried over. No code.
TL;DR: If your team already tests an API in Postman, the collection is most of an MCP server. Export it, import it into AnythingMCP, and every request becomes a tool that Claude, ChatGPT, Copilot and Cursor can call, with folders, auth and variables carried over. No spec, no code.
| Facts | |
|---|---|
| Format | Postman Collection v2.1 (JSON), by URL or pasted |
| One tool per | Request; folders become a name prefix |
| Carried over | Nested folders, auth at collection/folder/request level, query parameters, headers, raw JSON / form-data / urlencoded bodies, {{variables}} |
| Not carried over | Pre-request and test scripts (they are noted in the tool description, not run) |
| Hosting | Self-hosted with Docker (open source) or AnythingMCP Cloud |
| Last verified | 2026-09-26 |
Step 1 — Export the collection
In Postman: ⋯ → Export → Collection v2.1. Or host the exported JSON anywhere AnythingMCP can fetch it and import by URL. Environments are not part of the export; note the variables your requests need.
Step 2 — Run AnythingMCP
mkdir anythingmcp && cd anythingmcp
curl -fsSLo docker-compose.yml \
https://raw.githubusercontent.com/HelpCode-ai/anythingmcp/main/docker-compose.quickstart.yml
printf 'JWT_SECRET=%s\nENCRYPTION_KEY=%s\n' "$(openssl rand -hex 32)" "$(openssl rand -hex 32)" > .env
docker compose up -d # → http://localhost:3000
Or use AnythingMCP Cloud for an API that is reachable from the internet.
Step 3 — Create a REST connector and import
Create a REST connector with the API's base URL, then Import → Postman and paste the collection or its URL. Through the API:
curl -s http://localhost:4000/api/connectors/$CONNECTOR_ID/import \
-H "Authorization: Bearer $TOKEN" -H 'Content-Type: application/json' \
-d '{"source": "postman", "url": "https://example.com/exports/orders-api.postman_collection.json"}'
Step 4 — Fill the variables and tidy the tools
- Variables defined in the collection are resolved at call time; set their values (tokens, tenant IDs, base paths) as connector variables, where they are stored encrypted and hidden from the model.
- Path variables that are not defined in the collection (
/orders/{{orderId}}) become tool parameters the model fills in. - Rename and describe the tools: a request called "Get orders (copy)" is a poor tool name. The model picks tools by name and description.
- Pre-request scripts do not run. If a request depends on one (to sign it or fetch a token), use the connector's auth settings instead: OAuth 2.0, HMAC signing or a login-token flow cover most cases.
Step 5 — Connect your AI client
Add the MCP server URL to Claude (Customize → Connectors, step by step), ChatGPT, Cursor or VS Code.
How requests become tools
| Postman | MCP tool |
|---|---|
| Request "List orders" in folder "Sales" | tool sales_list_orders |
| Query parameters and headers | input properties |
| Raw JSON, form-data, urlencoded body | input properties mapped into the body |
| Auth on the collection, folder or request | the tool's auth, most specific wins |
| Request description | tool description |
Example prompts
- "Which orders came in through the web shop today?"
- "Look up customer 4711 and show their open tickets."
- "Create a test user in the staging environment." (a POST, if the role allows it)
FAQ
How do I turn a Postman collection into an MCP server? Export it as Collection v2.1, import it into a REST connector in AnythingMCP, and add the MCP server URL to your AI client. Each request becomes a tool.
Do Postman environments work? The export does not contain them. Set the values your requests need as connector variables; collection variables with the same names are resolved from there.
What about requests that need a pre-request script? Scripts are not executed. Configure the auth on the connector instead (OAuth 2.0, HMAC, API key, bearer, Basic or a login-token flow).
Is a Postman collection better than an OpenAPI spec? A spec is more precise about parameters and types, so prefer OpenAPI to MCP when you have one. A collection is the fastest route when you do not.
Can I limit what the AI may call? Yes. Give the MCP server a role that whitelists only the tools you want; the others are invisible to that client.
Related
Was this guide helpful?
