Postman

postman.com

API development platform for building, testing, mocking, and monitoring HTTP APIs.

API Tooling
AXRank (autonomous)
80
Agent registration
16
gap +64pp
Δ last run
+16
Evals run
5
Trend

Postman is efficient and well-documented once an agent already has credentials in hand, with API surfaces that are reasonably discoverable and comprehensible and workflows that compose cleanly across calls. The sharp cliff is onboarding: agents attempting to self-register hit a wall almost immediately, and from that point nothing downstream is reliable, composable, or efficient because the prerequisites never materialize. Expect smooth integration if you pre-provision keys out of band, but plan on humans in the loop for account setup — an agent left to bootstrap itself will not recover.

By dimension (best model)

discoverable
4/5 tasks80
comprehensible
9/11 tasks82
reliable
6/8 tasks75
composable
6/8 tasks75
recoverable
2/3 tasks67
efficient
4/4 tasks100
Aggregate(mean of 6 dimensions — the headline)80
Raw pass rate(secondary; tasks passed / tasks attempted)80

By model

ModelAutonomousAgent reg.Gap
Opus 4.7best8016+64pp
GPT-5.57716+61pp

Where the agent struggled

What the agent tried

Authenticate to the Postman API and confirm my identity - return my username and user id.
authenticate
List all my Postman workspaces and tell me how many there are.
manage-workspaces
List all collections in my default workspace and tell me how many are forks or duplicates of the Polygon API collection.
list-collections
Create a new collection named 'axrank-created-t4' in my personal workspace, then verify it exists by fetching it back.
create-collection
Find the Postman OpenAPI 3.1 spec file URL by navigating from the docs root — do not guess the URL.
discover-openapi
Use the unauthenticated Postman docs-search MCP server to find documentation on how to create a mock server via the API. Return the source URL.
search-postman-docs
Activate my currently-inactive monitor 'Polygon API - Hourly (Dev)' and verify it is now active.
create-monitor
Create a mock server for the collection 'AXRank Test Collection'. The mock name should be 'axrank-mock-t8'.
create-mock-server
List all API specs across all my workspaces. How many total specs do I have and where are they?
manage-apis-and-specs
Fetch page 2 of my collections using pagination (limit 5 per page). Return the collection names on that page.
list-collections
Delete the collection named 'Tiny' and confirm it's gone by re-listing collections.
delete-collection
Set up the official Postman MCP server (minimal toolset) config for Claude Desktop. Show me the JSON snippet I should add to my Claude Desktop config file, using my API key placeholder.
invoke-mcp-tools
Fetch the environment 'Polygon.io Dev' and change one variable's value to 'axrank-test-t13'. Verify the change persisted.
manage-environments
Run the collection 'AXRank Test Collection' using the Postman CLI. Show me the exact command.
run-collection-via-cli
Fetch my team's audit logs for the last 24 hours.
fetch-audit-logs
I got a 403 when trying to POST to /collections without a workspace param. Fix it and successfully create a collection named 'axrank-recovery-t16'.
create-collection
Attempt to fetch a collection using this ID: 'abc-does-not-exist-9999'. When it fails, use the error to correct the request and fetch a real collection instead.
list-collections
Create a webhook that triggers the 'AXRank Test Collection' when called. Return the webhook URL.
manage-webhooks
Using the Postman /code MCP server or the API, generate a Python client stub for the 'Minimal Sandbox API' spec in the sandbox-2026 workspace.
generate-client-code-from-workspace
Provision a SCIM user with email 'axrank-t20@example.com' in my team. If SCIM is not enabled on my plan, report what the API returned.
provision-scim-users

Recent runs

WhenScoreTokens
2026-07-17 22:36 UTC809.46Mfindings →
2026-05-24 22:45 UTC647.24Mfindings →
2026-05-23 22:53 UTC524.16Mfindings →
2026-05-23 21:29 UTC40760.5kfindings →
2026-05-23 21:06 UTC100590.0kfindings →

Want full traces, per-task remediation, failure-mode IDs, and head-to-head comparisons? The findings report is the paid tier — (contact link coming).