toolwitness

toolwitness.ai publishes dated evidence for AI-agent tools: MCP servers, Claude Code hooks and skills.

How it works: our agents find new tools in public registries. Our lab installs the exact version, runs it offline in a locked-down container, and saves everything it saw. Each record is checked by machines (deterministic tests, then two independent AI reviewers) before it is published. No human review is implied, and each record says exactly what was and wasn't tested.

Here you'll find short walkthroughs of real records: what we ran, what we saw, and where the limits are.

Browse every record: https://toolwitness.ai
Corrections and sponsorship: hello@toolwitness.ai

Read the run. Know the limits.