Goodeye MCP server

goodeye

Goodeye is a private home for the skills your AI follows and the verifiers its work must pass.

goodeye is the command-line surface for Goodeye. The same capability reaches your agent over an MCP server and a REST API too, so it works wherever your agent runs. This CLI talks to the public /v1 REST API.

How it works

A skill is a markdown runbook your agent follows. A verifier is a check its output has to clear. Goodeye keeps both in one account you own, and three things follow from that.

Private by default. Nothing is public until you publish a template, which is a separate step you take on purpose. To share without going public, grant a named user or team access, and the verifiers the skill references go with it at the same version. Revoking works the same way.

Always in sync. goodeye skills sync mirrors your hosted skills into the directories your tools already read, so Claude Code, Codex, Cursor, and anything else reading skill files from disk run the same current version. Configuring a target is all it takes: automatic sync is on from then on, so edit a skill once and every machine picks it up on its own.

Verifiers, hosted. A verifier can be deterministic (format, schema, tests, numeric bounds) or an LLM judge for the calls no test can make, like tone or image quality. Deploy a semantic verifier once, and every skill that references it runs that exact version, on your laptop, in CI, or on the machine of someone you granted it to.

The full picture is at https://goodeye.dev/docs/overview.

Your AI agent is the primary caller

The CLI is built to be driven by an AI coding agent on your behalf, though every command also works when you run it yourself. The intended loop: you ask your agent to run a Goodeye skill, it fetches the body with goodeye skills get (or goodeye templates get @handle/slug), and it executes those instructions as your runbook instead of printing or summarizing them. The get commands wrap the body in agent-facing markers so the calling agent knows to run it. Pass --output PATH or --json when you want the raw content instead.

Install

Requires Python 3.12 or later.

uv tool install goodeye
# or: pipx install goodeye
# or: pip install goodeye

The goodeye command is then on your PATH. Run goodeye update to upgrade to the latest release, or goodeye update --check to see whether one is available.

Quickstart

Bring a skill you already have

A skill file on disk is a directory holding a SKILL.md plus optional siblings, which is exactly what publish expects. Importing one is a single command.

goodeye login                                     # or: goodeye register (same browser sign-in)
goodeye skills publish ~/.claude/skills/my-skill  # also ~/.agents/skills, ~/.cursor/skills, or any path
goodeye skills list

It lands as a private hosted skill. Nothing is public until you publish a template, which is a separate step.

Sync it to every machine and agent

Point Goodeye at the directories your tools read, then pull:

goodeye skills sync target add --preset claude    # ~/.claude/skills
goodeye skills sync target add --preset agents    # ~/.agents/skills
goodeye skills sync target add --preset cursor    # ~/.cursor/skills
goodeye skills sync

Run the same commands on your other machines and they all read the current version. Once you have a target, automatic sync is on, so they stay that way without a second command: it only pulls new and updated skills, never overwrites local edits, and reports a conflict rather than clobbering it. Turn it off with goodeye skills sync auto off and it stays off.

Shortened. The full README is on GitHub.