Experimentation API for agents

Let your agents run real A/B tests.

p001 AB handles the bucketing, exposure logging, event tracking and statistics. Your coding or product agent creates the experiment, reads the results and ships the winner.

Get an API key llms.txt

Your agent sets up the test

Over MCP or REST it sizes the test, writes a hypothesis, defines variants with payloads and picks the conversion event.

We handle bucketing

Deterministic SHA-256 bucketing per unit: sticky assignments, weighted splits, traffic ramp-up, and exposures logged on first assign.

Track and decide

Send conversion and value events. Results come with lift, confidence intervals, p-values, SRM checks and a plain-language recommendation.

Connect an agent over MCP

Nine tools: create_experiment, assign, track_event, get_results, sample_size and more.

Claude Code

claude mcp add --transport http p001-ab https://ab.p001.ai/api/mcp --header "Authorization: Bearer ab_sk_…"

Or plain HTTP

Public keys can assign and track from browsers; secret keys do everything.

Assign a unit

curl -X POST https://ab.p001.ai/api/v1/assign \
  -H "Authorization: Bearer ab_sk_…" -H "Content-Type: application/json" \
  -d '{"unitId": "user_123", "experiments": ["checkout-cta"]}'