Skip to content

Commands

After postman-mcp init, seven slash commands are available inside Claude Code. The sync commands aren't separate implementations: they all go through the same validate → verify → diff → confirm → write pipeline, and differ only in which routes Claude analyzes and where the result lands.

Command One-liner
/postman:syncapi Sync one route. The kernel that everything else is built on.
/postman:syncchanges Sync what changed since the last sync. The one you'll run most.
/postman:sync Sync everything in one file, module, or directory.
/postman:syncall Sync the whole codebase. Usually a first-run or post-refactor thing.
/postman:prompt Sync from a plain-English instruction (add error responses, headers, etc.).
/postman:createenv Generate a Postman environment from your code.
/postman:status Show drift without writing anything.

The diff-then-confirm contract

Every write-capable command follows the same two-phase contract, so nothing reaches Postman as a surprise:

sequenceDiagram
    participant You
    participant CC as Claude Code
    participant S as MCP server
    participant P as Postman
    You->>CC: /postman:syncapi create_payment
    Note over CC: Claude reads the code and authors
collection.json + metadata.json
    CC->>S: sync_files(confirm=false)
    S->>P: GET collection (read structure)
    S-->>CC: validated, verified diff preview (writes nothing)
    CC-->>You: diff ... Write to Postman? [y/n]
    You->>CC: y
    CC->>S: sync_files(confirm=true)
    S->>P: PUT merged collection
    S-->>CC: written

On n, nothing is written. There's no flag to skip this step; see the safety rules. Once the result is shown (either the write confirmation or the n abort), the command ends. Claude doesn't keep going with more analysis or commentary after that.

Natural-language sync with /postman:prompt

For free-form instructions — "add error responses to the payments routes", "give every endpoint an X-Request-Id header", "rewrite the login description" — use /postman:prompt "<text>". Claude reads the instruction, picks the right scope and target, and folds the "how" directly into the collection it authors. The instruction is read by Claude, not by the MCP server, which has no prompt parameter and stays deterministic; everything still goes through the same diff-then-confirm gate. See the Prompt & skill layer.

Terminal vs. Claude Code

The /postman:* commands above run inside Claude Code. Setup commands run in the terminal and aren't slash commands:

  • postman-mcp init: one-time project setup.
  • postman-mcp doctor: re-validate the whole setup chain.
  • postman-mcp serve: boot the MCP server (Claude Code launches this for you, you shouldn't need to run it by hand).
  • postman-mcp version: print the version.