OpenCap.dev - Screen-recording for agents

Loom, but for agents Your coding agent runs `opencap record start`, drives the demo (Playwright, the app, whatever), runs `opencap record stop`, and gets a share URL paired with a structured event log of every action it took. Then it drops the link wherever proof-of-work belongs: nPR description, Linear ticket, Slack channel. CLI + MCP first. Built for Claude Code, Codex, Cursor. Free forever for CLI + MCP.

OpenCap exists because a few weeks ago I asked Claude Code to record a Loom of a feature it just shipped for a client, post the replay link to Linear and the PR, and DM the client in Slack. To my surprise, no Loom CLI exists. The whole stack assumed a human was clicking record. So Claude Code did the next best thing: Playwright screenshots, dumped to markdown, converted to a 20-page PDF the client probably never opened. That afternoon I started building OpenCap. Three things surprised me through the build: 1. The event log matters more than the video. The structured JSONL stream alongside the recording lets you trim by natural language ("cut to where the bug was reproduced") instead of scrubbing the timeline. 2. The share-link distribution is the real product. The recording is the artifact, but the agent dropping the link into Linear, Slack, and the PR without a human relay is what saves time on every ship. 3. MCP is a forcing function. Designing the surface as MCP tools first made every UX decision easier, because the user is an LLM, not a human. Free forever for CLI and MCP. Paid tiers gate longer recordings and password-protected shares. Open to all feedback

OpenCap.dev - Screen-recording for agents

Replies