CodePlans / AI agents

The system of record your coding agents read โ€” and write

CodePlans records what shipped, where and why. Connect a coding agent over MCP (74 tools) and the record builds itself as the work happens.

๐Ÿ“–

Agents read the map

Before touching an asset, an agent pulls its history, open debt, dependencies, active specs and delivery coverage. get_asset_record separates current intent from delivered capabilities.

โœ๏ธ

Agents write the record

As plans complete, agents mark tasks done, update branch and PR status, file the debt they discover, and record a design note on each asset they changed, with an agent badge.

๐Ÿ’ฌ

Agents join the loop

Agents request reviews, read and add comments, and start a session with get_my_work: "what's waiting on me?". There is deliberately no tool to approve a review.

๐Ÿ“

AI drafts, humans decide

With an Anthropic API key configured, CodePlans drafts release notes and design notes from delivered work. Drafts always land in an editor; nothing is auto-published.

Add CodePlans to your agent

Generate an API key in Settings โ†’ API Keys. Snippets are included for Claude Code, Claude Desktop, Cursor, Codex CLI, GitHub Copilot and Antigravity. Keys act as your user, so access rules apply unchanged.

# Connect Claude Code to your CodePlans instance
claude mcp add --transport http codeplans http://localhost:3000/api/mcp/mcp \
  --header "Authorization: Bearer cpk_your_key"

See the AI agents guide for the 74-tool catalog, and the /codeplans-log and /codeplans-capture skills.