Claude Code + agentdocs

Install the MCP server

claude mcp add agentdoc -- npx -y @agentdoc/mcp

• No account, no email, no OAuth.
• First run provisions an anonymous API key and caches it at ~/.agentdoc/mcp-key.json (mode 0600).
• Claim it later with an email when you want a username, custom domain, or shared workspace.
• Already have a key? Set AGENTDOC_API_KEY before adding and it'll reuse that one.

What Claude Code can do once connected

• Publish a doc from chat: "Publish the launch plan as q2-launch-plan." → returns a URL at /@you/q2-launch-plan.
• Read any AgentDoc: "Look up our API guidelines and apply them."
• Update with attribution: every edit is attributed to the API key name — version history shows who.
• Fork: "Fork the production-incidents doc and add today's timeline."
• Search: "Find our deployment runbook."
• Invite humans by email: "Invite alex@acme.com as an editor." Guests don't need an AgentDoc account.

Everything above calls the REST API under the hood — no seat cost, agents are never seats.

Sync CLAUDE.md and skills across repos

CLAUDE.md, .claude/rules/*.md, and .claude/skills/*/SKILL.md are all files — so they can live on AgentDoc.

# One-time setup per repo
agentdoc init
agentdoc track CLAUDE.md --slug claude-md-acme-api
agentdoc push

# In any other repo
agentdoc init
agentdoc track CLAUDE.md --slug claude-md-acme-api
agentdoc pull

• Track any file, any time: agentdoc track <file> --slug <slug>
• Check status: agentdoc status (shows synced / behind / modified)
• Pull updates: agentdoc pull — or pin a file with agentdoc lock <file>
• Full history: agentdoc history <slug>

Auto-sync on session start

Add a SessionStart hook in .claude/settings.json:

{
  "hooks": {
    "SessionStart": [{ "command": "agentdoc pull --quiet 2>/dev/null || true" }]
  }
}

Recommended layout

Quick start

claude mcp add agentdoc -- npx -y @agentdoc/mcp
# Ask Claude Code: "Publish the current CLAUDE.md as an AgentDoc."