Assay

The CI/CD Standard for Agentic Systems

Assay is a strict Policy-as-Code engine for Model Context Protocol (MCP). It validates that your AI Agents use tools correctly, enforcing schema limits and sequence rules before they hit production.

What is Assay?¶

Install Now

Get the binary in seconds via our new installer.

getassay.dev
For Vibecoders

You build agents with natural language. Assay is your Guardrail. Connect your traces, run assay validate, and see if your agent is trying to delete the production database.

Python Quickstart
For Engineers

You need Determinism. Assay is a high-performance Rust binary that enforces rigid JSON Schemas and sequence constraints in CI. No flaky evals.

CLI Reference

How it Works¶

1. Initialize¶

Run the wizard to auto-detect your project type and generate secure defaults.

assay init

2. Capture Traces¶

Log your agent's MCP tool calls to a JSONL file.

{"tool": "run_tests", "args": {}}
{"tool": "deploy_service", "args": {"env": "prod"}}

3. Validate¶

Run the validation engine (Stateless). Supports SARIF for GitHub Advanced Security.

assay validate --trace-file traces.jsonl --format sarif

Result	Status	Output
Pass	✅	`exit code 0`
Fail	❌	`exit code 1` + SARIF report
Error	⚠️	`exit code 2` (Config/Schema validation)

Key Features¶

Stateless: No database required. Validate in GitHub Actions, GitLab CI, or local pytest.
The Doctor: assay doctor automatically diagnoses config errors.
Agentic Contract: JSON output optimized for AI agents (--format json).
CI-Native: assay init --ci generates GitHub Actions workflows.
Fast: Written in Rust. <10ms overhead.