Happy slice of mcp pizzaMCP.pizza
eval-view

eval-view

MCP.Pizza Chef: hidai25

EvalView records what your agent does today — which tools it reaches for, in what order, and what comes back — then compares every later run against that recording. When a model update or a small wording change makes it skip a step or pick the wrong tool, you find out before your users do. The basic comparison runs on your machine with no key at all; add an OpenAI key only if you also want the answers scored for quality.

Coding
Data

Use This MCP server To

Record what my agent does today as a baseline See exactly which step changed after a model update Block a change that quietly breaks my agent Compare two versions of a prompt side by side Get a summary of what changed on every pull request Try it on a demo agent in thirty seconds

README

EvalView
Snapshot testing for AI agents.
Record what your agent does today. Get told when it silently changes.

PyPI version PyPI downloads Daily dogfood GitHub stars License

Your agent returns 200 and looks fine. But a model update, a provider change, or a one-line prompt edit just made it skip a clarification, call the wrong tool, or quietly drop output quality. Your tests still pass. Your users notice before you do.

EvalView snapshots your agent's behavior — the tools it calls, in what order, with what output — and tells you the moment that behavior changes. Like Jest snapshots, but for tool-calling, multi-turn agents.

demo.mp4

↑ 30-second live demo — no API key needed

Quick Start

pip install evalview
evalview snapshot    # Record your agent's current behavior as the baseline
evalview check       # After any change, diff against the baseline

That's the whole loop. check returns one of:

  ✓ login-flow        PASSED          behavior matches baseline
  ⚠ refund-request    TOOLS_CHANGED   called a different tool, or in a different order
  ✗ billing-dispute   REGRESSION      score dropped — output quality fell

eval-view FAQ

Do I need a paid key?
No — the tool and order comparison runs offline; a key is only needed for quality scoring.
How do I install it?
One pip install command, then two commands: one to record a baseline, one to check against it.
Which agent frameworks does it work with?
LangGraph, CrewAI, OpenAI, Claude, Mistral, Ollama, MCP-based agents, and anything reachable over the web.
Can I use this to stop a bad change reaching production?
Yes — it runs on every pull request and fails the check when behaviour drifts from the recording.
What if my agent is not fully predictable?
You can record up to five acceptable paths, so normal variation does not get flagged.
Can I try it without wiring up my own agent?
Yes — a built-in demo walks through the whole record-and-compare loop in about thirty seconds.