# Parakh

> Agent writes the check → secrets never touch the model → proof the PR can trust → then get out of the way.

Parakh is a short-lived check runner for coding agents. Collections and workspaces stay in Postman, Bruno, or your repo. Parakh runs isolated checks against real APIs and browsers, returns shareable PR reports (API traces, replayable curl, screenshots), then discards the check.

## What Parakh is

- A safe runner for checks your coding agent writes
- Credential storage outside the model; secrets never enter chat
- Ephemeral isolated workers; the check dies after the run
- PR-ready reports: written summary, API traces, replayable curl, screenshots when needed
- MCP-first interface for Claude Code, Codex, Cursor, and other agents
- HTTP API for scripts and automation

## What Parakh is not

- Not a Postman or Bruno replacement; keep collections where they are
- Not an always-on monitor or synthetic uptime product
- Not an AI that writes and maintains your test suite for you
- Not a dashboard product; the agent is the interface

## Surfaces

- Homepage: https://parakh.prashamhtrivedi.app/
- HTTP API docs (Scalar): https://parakh.prashamhtrivedi.app/api/docs
- OpenAPI: https://parakh.prashamhtrivedi.app/api/openapi.json
- MCP: https://parakh.prashamhtrivedi.app/mcp (stateless, POST only)

## Key claims

- Secrets never enter the model
- Proof reviewers can trust in the pull request
- Checks are ephemeral; setup stays, the run does not