Percia for AI agents

Point it at a camera, describe the moment, get a signed webhook.

Percia is built so a coding agent can use it end to end, without a person clicking through a browser.

How an agent starts

Every docs page and this page also come as markdown: send Accept: text/markdown, or fetch /coding-agents.md and /docs/<page>.md.

Connect your agent

Create an API key in the app under Settings > API keys and give it only the permissions the job needs. Put it in PERCIA_API_KEY; never paste it into a repository.

Claude Code

claude mcp add --transport http percia https://playground.percia.ai/mcp --header "Authorization: Bearer $PERCIA_API_KEY"

Without a key, sign in with OAuth instead. Run this, then /mcp inside Claude Code and choose Authenticate:

claude mcp add --transport http percia https://playground.percia.ai/mcp

Claude.ai

Settings > Connectors > Add custom connector. Name it Percia and use this URL. You sign in and choose what to allow:

https://playground.percia.ai/mcp

Cursor

Add this to .cursor/mcp.json (project) or ~/.cursor/mcp.json (all projects):

{
  "mcpServers": {
    "percia": {
      "url": "https://playground.percia.ai/mcp",
      "headers": { "Authorization": "Bearer ${env:PERCIA_API_KEY}" }
    }
  }
}

Docs only (no key)

claude mcp add --transport http percia-docs https://playground.percia.ai/mcp/docs

Cursor and Claude.ai take the same URL. It also offers the task skills as skill://<name>/SKILL.md resources.

Skills

Short task skills for coding agents: connect a camera, create a watch, build an agent, set up a webhook, debug a delivery, review events, plus the integrator rules in percia. Each is a plain file at /skills/<name>/SKILL.md, for example /skills/percia/SKILL.md. The agent skill is not published to a public repository yet. This is its install line from the docs:

npx skills add <owner>/percia-cf --skill percia

Published pass rates

We run real coding agents against a local Percia server. Each run is one task in one mode, and a checker reads the server's state (not just the agent's reply) to confirm the outcome. There is one task per dashboard capability: create a watch, pause/edit/resume, delete with and without permission, events and keyframes, agents, sites, phone cameras and plan usage. The modes are how the agent reaches Percia: code mode (/mcp), all tools (/mcp?tools=all), docs only (public docs and curl) and skill only (the skills/ folder and curl).

Loading results…