Claude Code plugin · MCP server

A git-committed kanban board your AI agents run and remember.

The board is the database. Coordination, work-in-progress locks, and durable learnings — all committed markdown in your repo.

Install View on GitHub

a finding, promoted through the state machine

needs: tdd

needs: review

needs: validate

resolved

Why it's different

Competitors force a choice: a board that's visible but dumb, or coordination that's smart but hidden in a database. engineering-board refuses the trade-off.

Visible, diffable state

Your agents' board is committed markdown — cards, an index, a graph — reviewed in the same PRs as code.

engineering-board/ · BOARD.md · GRAPH.yml

Durable memory

Recurring lessons promote into committed Learning entries (L###) that survive session boundaries — surfaced at session start and in a dedicated Learnings panel on the board view.

learnings-curator → learnings/L###

Collision-free agents

Atomic claim-locking lets multiple worker agents run in parallel without stepping on each other.

board-claim-acquire.sh · tested

Autonomous pipeline

Findings flow through a tdd → review → validate state machine, driven by the Stop hook.

tdd-builder · code-reviewer · validator

Runs where you are — and everywhere else

A native Claude Code plugin and an MCP server (11 tools) over the same board format.

.mcp.json · protocol 2025-06-18

Nobody else occupies the intersection

Honest comparison. Every competitor is linked; traction figures are snapshots.

Tool	Git-committed board	Durable memory	Atomic multi-agent locking	Claude-native	MCP
engineering-board	Yes	Yes	Yes	Yes	Yes
Backlog.md	Yes	No	No	CLI	Yes
Agent-MCP	RAG DB	Yes	Yes	No	Yes
kanban-mcp	SQLite	No	No	No	Yes
claude-code-workflows	Ephemeral	No	No	Yes	No
Flux	Side branch	No	No	No	Yes

Where they're better: Backlog.md is the category leader (~5.9k★) with a polished standalone CLI and TUI; Agent-MCP ships a live dashboard. engineering-board answers visualization with /board-view — a committed, offline, zero-dep HTML board — see this repo's own live board, republished on every merge (no daemon, on-brand: the view is just another in-repo projection) — and bets on the four-way intersection, not any single feature.

Install both ways

One shared board format. Two thin adapters.

Claude Code plugin

/plugin marketplace add GhostlyGawd/engineering-board
/plugin install engineering-board

# then, in a project:
/board-init my-project

Installing the plugin also registers the MCP server via the bundled .mcp.json.

MCP server (any client)

# Claude Code
claude mcp add engineering-board -- \
  python3 /path/to/engineering-board/mcp-server/engineering_board_mcp.py

# Claude Desktop: claude_desktop_config.json
"mcpServers": {
  "engineering-board": {
    "command": "python3",
    "args": ["/path/to/mcp-server/engineering_board_mcp.py"]
  }
}

Pure python3, zero dependencies. 11 tools over stdio.