Autonomous onboarding
Machine-readable discovery tells the agent where to fetch its skill, prove ownership, and connect the MCP host.
MCP-native experimentation infrastructure
It proves domain control, receives scoped credentials, creates and ramps experiments, ships the browser SDK, and reads the result through MCP.
Built for an AI agent, not another dashboard.
https://aibtest.ai-neos.com/mcp
One trusted connection
The credential exchange stays inside the MCP host. The agent sees tools and outcomes, never the private key or bearer token.
The agent installs the trusted skill and asks AIBTest to register the exact hostname.
It publishes an HTTPS challenge file or DNS TXT record using the access you approved.
It defines variants, traffic, selectors, a primary metric, and a stopping rule through MCP.
The browser SDK holds assignment, records exposure and conversion, then the agent reads the report.
Credential-safe by design
A local CLI proxy creates signed client assertions and renews short-lived tokens. Credentials stay outside the model context while MCP tools remain fully usable.
$ aibtest connect --server \
https://aibtest.ai-neos.com example.com
verification https_file ready
credential stored by local host
$ aibtest serve
mcp_url http://127.0.0.1:43179/mcp
✓ agent ready to create experiments
A complete operating surface
Machine-readable discovery tells the agent where to fetch its skill, prove ownership, and connect the MCP host.
Sticky allocation, traffic from 1–100%, anti-flicker guidance, and a deliberately narrow DOM mutation surface.
Successful writes for exposure and conversion, descriptive reports, quota headers, and an agent issue channel.
Start from discovery
https://aibtest.ai-neos.com/.well-known/aibtest