Test and debug APIs, MCP servers and AI systems in one place

REST, GraphQL, WebSocket, MCP, LLMs, RAG and agents — with a test runner, evaluation lab and trace viewer. Local-first, private, and the same engine in your CI.

Download

View on GitHub

Read the docs

REST & GraphQL client

Requests with every common auth type, bodies, cookies, scripts and assertions. GraphQL introspection with schema-aware autocomplete. Large responses stream to disk.

API testing

MCP inspector

Connect over stdio, Streamable HTTP or SSE. Discover tools, resources and prompts, run tools from generated forms, and trace every JSON-RPC message.

MCP debugging

AI Lab

Test prompts on OpenAI-compatible, Azure, Anthropic, Gemini, Amazon Bedrock and Ollama models — or an offline mock. Streaming, time to first token, tokens, cost and side-by-side comparison.

AI testing

Evaluations

Stream datasets through deterministic, semantic and LLM-as-judge evaluators, including RAG, agent tool-use and safety checks. Compare runs against baselines.

Evaluation guide

Test runner & CI

YAML test suites with parallel workers, retries, dependencies and resumable runs. Save any request as a test, run it in CI from the run itself. JUnit, JSON, HTML and Markdown reports, and the testpion CLI for any CI system.

Test runner

Visual flow designer

Build multi-step API flows on a canvas — requests, conditions, loops, polling, sub-flows — debug them with breakpoints, run them per data row, per environment or under load, and deploy them to a schedule, a webhook, an MCP tool or CI. The file stays plain YAML.

Design a flow

Monitors & certificates

Run collections or flows on a schedule with alerts, response-time limits, uptime by day and per-request numbers. Every HTTPS response records its certificate, so an expiring one never surprises you.

Monitors

Charts that explain

A Home dashboard with what needs attention, run charts down to DNS, TCP, TLS and server time, each test's history and the flaky ones, load and monitor trends, and usage of MCP tools and AI models.

All features

Built for AI agents

testpion mcp-server gives agents more than 120 tools, from sending requests and designing flows to what needs attention, with secrets redacted and production guarded. testpion agents init writes AGENTS.md, every CLI command's --json is one envelope, and the docs publish llms.txt.

MCP server

Local-first and private

No account and no telemetry. Secrets are encrypted with your OS keychain and redacted from logs, traces and reports. Scripts run in a WebAssembly sandbox.

Privacy

See it in action

TestPion sending a GET request from the Veterinary API collection, showing a 200 response as a JSON tree and three passing assertions — A REST request from a collection, with the JSON response and passing assertions.

The MCP inspector connected to a stdio server, running the search_customer tool from a form generated from its JSON Schema — MCP inspector: run a tool from its schema.

A test run of 22 REST, GraphQL, MCP, LLM, RAG and agent tests, all passing, with evaluation score cards — One run across REST, GraphQL, MCP, LLM, RAG and agents.

The AI Lab playground with a prompt template, streamed JSON output, latency, time to first token and token counts — AI Lab: prompt, stream and measure.

A trace waterfall of an agent test with LLM steps, an MCP tool call and evaluation — Traces for every request, tool call and model call.

See all features

Runs on your operating system

TestPion is a desktop app for Windows, macOS and Linux, and every release has installers for each:

System	Installers
Windows 10 and 11, x64	Installer (`.exe`) or a portable `.exe` that runs without installing
macOS 12+	`.dmg` for Apple Silicon and for Intel
Linux x86_64	AppImage, `.deb` and `.rpm`

The testpion command-line runner uses the same engine and runs anywhere Node.js 22+ runs — see installing the CLI.

Download TestPion · Installation guides

Local-first by design

Your data stays on your machine. Workspaces are plain files (collections, environments, YAML tests) that you can commit to git. There's no account and no cloud backend.
Secrets never touch workspace files. Tokens and API keys are encrypted with Windows DPAPI, the macOS Keychain or Linux Secret Service, and redacted from every log, trace, report and export.
No telemetry. Requests and prompts go only to the servers and AI providers you call.
Deterministic where it matters. AI-judge and heuristic scores are always labelled, so they're never confused with real assertions.

Read the privacy notes.

Latest release

TestPion 0.53.1, released 2026-10-19. See what's new or all release files on GitHub.

Developed in public on GitHub

The source code, issues and releases are public at github.com/nasimuddin-dev/testpion. Report a bug or request a feature, browse releases, or read the source.