Honeypot MCP server

Honeypot

A free index of skills for AI agents. Machine-readable. No sign-up.

Live: https://honeypot-e6c.pages.dev

Fetch it by machine

The index is one JSON file. schema_version is "1"; entries live under .skills.

# everything
curl -sL https://honeypot-e6c.pages.dev/api/skills.json | jq '.skills[] | .name'

# one category only (cheaper than filtering client-side)
curl -sL https://honeypot-e6c.pages.dev/api/by-category/trading.json | jq '.skills[] | .name'

# install one skill
curl -sL https://honeypot-e6c.pages.dev/api/skills.json \
  | jq -r '.skills[] | select(.name=="kalshi") | .install'
# then run the command it prints

Linkable searches

The site takes the same filters as URL params, so any view can be shared or bookmarked:

• ?q= — ranked match on name, description, and tags (a one-character typo still matches; hits are highlighted)
• ?tag= — exact tag
• ?category= — exact category

Params combine: https://honeypot-e6c.pages.dev/?category=trading&tag=prediction-markets. Filtering happens in the browser, so curl on that URL returns the unfiltered page. The page ranks matches and tolerates a one-character typo. A script can apply the same category and tag filters, plus a plain substring check, to the JSON:

# substring stand-in for ?q= (the page ranks; this check does not)
curl -sL https://honeypot-e6c.pages.dev/api/skills.json | jq '.skills[]
  | select(.category=="data")
  | select(.tags | index("sports-data"))
  | select((.name+" "+.description+" "+(.tags|join(" "))) | ascii_downcase | contains("odds"))
  | .name'

Each entry has name, description, category, tags, last_verified (date), skill_md_url, and install. New additions: /api/changelog.json and /feed.xml (RSS).

Editor's picks live at /api/editors-picks.json. There's no score: a skill is a pick when it passes every check in PICK_CRITERIA in scripts/gen.py (description length, tags, verified recently, SKILL.md link). The file carries the checklist itself as criteria and criteria_text, and the site prints that text as-is.

Every card on the site shows its last_verified date. Anything not re-verified in STALE_AFTER_DAYS (180, set in scripts/gen.py and published as stale_after_days in /api/editors-picks.json) gets a "stale" badge and drops out of the picks. Freshness is date-based only; nothing pings the skills. The same JSON is mirrored at https://raw.githubusercontent.com/oidsdev/honeypot/main/api/skills.json.

Submit a skill

1. Write your skill as a SKILL.md. Start from skill-template.md.
2. Fork this repo.
3. Add it at skills/<your-skill>/SKILL.md.
4. Lint it from the repo root:

python3 scripts/lint_skill.py skills/<your-skill>

pass is clean. fail prints file:line: reason for each problem: missing name or description, a description outside 20-300 characters, a banned phrase (the list is in scripts/lint_skill.py), or a relative link that does not resolve to a file. 5. Open a pull request. We merge fast. 6. Add the badge to your repo:

[![Indexed on Honeypot](https://honeypot-e6c.pages.dev/badge.svg)](https://honeypot-e6c.pages.dev)

Per-skill badges live at /badge/<skill>.svg. The image shows the skill name, a version, and a check when the skill passes the linter in scripts/gen.py.

[![cron on Honeypot](https://honeypot-e6c.pages.dev/badge/cron.svg)](https://honeypot-e6c.pages.dev/skills/cron/)

Version is the optional version field in the SKILL.md frontmatter. With no version, the badge shows last_verified. The check means the frontmatter has name and description, name matches the directory, and the index entry has a category, tags, and a last_verified date. The SVG is static and cached for a day.

Requirements (learned the hard way — PRs missing these get bounced):

• Frontmatter must include name and description. Directory CLIs (e.g. skills.sh) silently skip files without them.
• Pick one category (trading, data, identity, meta, tooling, or propose a new one in the PR) and add tags plus a current last_verified date — these feed the JSON index.
• MIT-compatible license only. We can't index what we can't redistribute.
• Spam, duplicates, and anything malicious get closed. Every listing is PR-gated and read before it merges.

Layout

• index.html — the site
• ja/index.html — Japanese homepage
• search.js — ranked search used by the site (no dependencies)
• categories/<name>/ — landing page for each of the largest categories (generated)
• api/skills.json — the machine-readable index
• scripts/gen.py — rebuilds everything derived from skills.json (run after any change)
• scripts/lint_skill.py — checks a submitted skill (frontmatter, description length, banned phrases, relative links)
• scripts/lint-fixtures/ — five sample skills the linter is checked against (--fixtures)
• llms.txt — agent-readable description
• skill-template.md — blank SKILL.md template
• skills/<name>/SKILL.md — the indexed skills
• tests/test_search.js — node tests/test_search.js
• tests/test_gen.py — python3 -m pytest tests/test_gen.py
• tests/test_mcp_server.py — python3 tests/test_mcp_server.py
• badge/<name>.svg — per-skill badge, written by scripts/gen.py
• _headers — cache header for /badge/* (one day)

License

MIT — Orbital Desk LLC.