MeshTest

Live API testing for the AI era.

Import a Postman collection or OpenAPI spec. One command later you have a regression-tested API with drift detection, baseline tracking, and a built-in MCP server so any AI agent — Claude, Cursor, Windsurf, your own — can drive your tests directly.

Works for any HTTP API: a monolith, a microservice, a Lambda behind API Gateway, a third-party SaaS dependency, or a GraphQL endpoint. If curl can hit it, MeshTest can test it.

What you get in 60 seconds

meshtest generate --from openapi spec.yaml    # auto-detected auth, CRUD chains, expectations
meshtest run --env staging                     # live tests against your real environment
meshtest diff --record --tag v1.2.0            # snapshot for drift detection

Three commands. No GUI. No SaaS account. No hand-written tests.

Two paths, one manifest

Path	Who runs it	Where
CLI (@meshtest/cli → meshtest binary)	Engineers, CI	Terminal, GitHub Actions, any container
MCP server (@meshtest/server → meshtest-server binary)	AI agents	Claude Desktop, Cursor, Windsurf, custom integrations

Same meshtest.yaml. Same tests. Same auth. Same baselines. Whether a human or an agent triggers the run, the result is identical.

What makes it different

• You don't write the manifest. meshtest generate ingests your Postman or OpenAPI source and produces a runnable manifest — including auth, request bodies, expected statuses, and CRUD capture chains for dynamic IDs.
• Drift detection by default. Every run can compare against a stored baseline. Status changed from 201 → 200? Body schema gained a field? Latency doubled? You find out before customers do.
• AI-native from the first commit. Five MCP tools. Claude Desktop verified. The agent doesn't need glue scripts — it speaks meshtest natively.
• No vendor lock-in. Manifests are plain YAML. Baselines are plain JSON. The runner is open. Move to another tool any time without rewriting your tests.

Why MeshTest exists →

Who it's for

• Backend engineers who maintain one or more HTTP APIs and need regression testing without standing up Datadog Synthetic or Postman Cloud.
• AI-tooling teams who want their agents to test production behaviour, not just read source code.
• Pre-launch and mid-launch products where every release ships breaking changes that "should work" — and sometimes don't.
• Teams already on Postman/OpenAPI who want their existing artefacts to do double duty as test specs.

Where to start

If you want to...	Read this
Run your first test in 5 minutes	Quickstart
Understand the manifest format	Your first manifest
Handle CRUD endpoints and dynamic IDs	Test data management
Set up authentication	Authentication
Wire it into AI agents	MCP integration
See every CLI flag	CLI reference
Compare to Postman / Newman / k6	Why MeshTest

Status

Pre-1.0. 147 tests passing across 4 packages. v0.1 is feature-complete — generate, run, diff, MCP server, full auth system, capture chains, setup/teardown, baseline secret-stripping, consumer-level meshtest.config.json with manifest auto-discovery, and meshtest init to scaffold a new project. v0.2 ships the remaining 5 MCP tools and (optional) cloud baseline storage.