How it works
Microfilm fits the way agents already test. Here's the full path — from connecting your agent to sharing verifiable evidence — step by step.
Point your AI coding agent — Claude Code, Cursor, whatever you run — at Microfilm’s Model Context Protocol server. There’s no desktop app to install and no browser extension; the server is the only thing the agent needs to reach.
As it runs tests, your agent records each event to the server — the procedure it ran, the work item it exercised, and the result — through a typed tool call. The evidence is captured at the moment of work, not reconstructed from a log afterward.
Every event is signed with a KMS-backed key (ES256) and linked into a tamper-evident reel. The agent gets back a record that can be verified offline — actor, content, and time fixed at the moment of capture.
Signed records flow into the web portal. The portal is passive — it captures nothing — and organizes every record by work item, result, agent, and reel.
Find any record, then share a scoped, read-only link. Because every record is signed and chained, the recipient verifies it offline — you’re handing over proof, not a claim.
A traceability view links each requirement to the signed records that prove it was tested, surfaces untested gaps, and exports as an audit package on demand — a view over your signed evidence, not a separate matrix to maintain.
Create an account, connect your agent to the MCP server, and write signed, verifiable evidence as it tests.