Skip to main content
There are two ways in. The fastest is to connect an MCP client and let your AI client do the driving; the classic path is to add the collector to your agent and build a mirror from its traces. They end in the same place: real sessions replaying against every change, in CI.

​
Option A: connect an MCP client

Mirrors hosts an MCP server that exposes the full product surface — build, explore, query, and eval mirrors from Claude Code, Cursor, VS Code, ChatGPT, Codex, and any other MCP client. Sign-in happens in the browser on first use; no API key needed. 
claude mcp add --transport http mirrors https://api.runmirrors.com/mcp
Then run /mcp → mirrors → Authenticate and ask your client to build a mirror from your traces. See MCP server for every client, one-click installs, and headless setups.

​
Option B: add the collector

1

Get an API key

Sign in at runmirrors.com and mint a workspace API key (mk_live_…) under Settings → API keys.
2

Add two lines to your agent

Install the collector and initialize it before your agent runs. It auto-instruments LangChain/LangGraph, Anthropic, and OpenAI, and ships traces in the background — non-blocking, and it never raises into your app.
# pip install mirrorkit
import mirrorkit
mirrorkit.init(api_key="mk_live_...", project="my-agent")

// npm install mirrorkit
import * as mirrorkit from 'mirrorkit';
mirrorkit.init({ apiKey: 'mk_live_...', project: 'my-agent' });

// go get github.com/ai-singhal/mirrorkit-go
mirrorkit.Init(mirrorkit.Options{APIKey: "mk_live_...", Project: "my-agent"})
Run your agent normally — traces start streaming. Details per language: Python, TypeScript, Go.
3

Build a mirror

From the dashboard (Ingest → Build), or from the CLI:
pip install "mirrorkit[cli]"
mirrors login                                   # paste your mk_live_… key
mirrors build --project my-agent --name my-agent
Mirrors turns the traces into a schema, a seeded database, and tool bindings — a runnable copy of your agent’s world, scored per tool for fidelity.
4

Replay and query it

mirrors query my-agent "cancel my flight"   # one-shot run + trace
mirrors chat my-agent                       # multi-turn conversation
Or use the Playground in the dashboard. Same session, two versions of the agent, one caught regression.
5

Gate merges in CI

In Quickstart → CI eval gate, connect the GitHub App and bind your repo to a mirror and an eval set. Every PR that touches your agent replays the eval set and reports a Mirrors eval gate status check you can make required. See CI gate.

​
Next steps

How it works

Fidelity scoring, deterministic seeding, business context, and evals.

CI gate

Block regressions at the pull request, Vercel-style.