agent-dashboard-mcp

An MCP server that exposes a self-hosted multi-agent dashboard's HTTP API as first-class tools. Instead of teaching every agent to curl an endpoint and juggle a bearer token, you connect them to this MCP server once and they get typed tools for memory, kanban, inter-agent messaging, daily logs, and bot status.

Built for a multi-agent assistant setup where several agents (a lead, a reviewer, a worker, ...) share a common SQLite-backed dashboard and need to read and write that shared state reliably.

Why an MCP server instead of curl

When agents call a REST API by shelling out to curl, they make avoidable mistakes: wrong header, forgotten token, malformed JSON, no schema. Wrapping the same API as MCP tools gives each call a name, typed arguments, and a docstring the model can read — so the agent picks the right tool with the right fields instead of reconstructing a request from memory.

Tools

Memory

memory_search(query, agent, tier, limit, mode) — hybrid semantic + keyword search
memory_save(content, agent, tier, keywords) — save to a hot/warm/cold/shared tier

Kanban

kanban_list(status, priority, assignee, limit)
kanban_create(title, description, status, priority, assignee)
kanban_update(card_id, ...) / kanban_move(card_id, status, sort_order)
kanban_archive(card_id)
kanban_comment(card_id, author, content) / kanban_comments_list(card_id)

Inter-agent messaging

message_send(to, content, from_agent)
messages_list(agent, status, limit)
message_mark_done(message_id, result)

Daily log & agents

daily_log_append(content, agent)
agents_list()

Bot ops (local)

bot_tmux_status(session_name) — is a tmux session alive
bot_log_tail(log_path, lines) — tail a bot's log file

Configure

Copy .env.example to .env and set:

DASHBOARD_API_URL=http://localhost:3420   # your dashboard API base URL
DASHBOARD_TOKEN=...                        # bearer token, or:
DASHBOARD_TOKEN_PATH=.dashboard-token      # path to a file with the token

The token is read once at startup (env first, then file fallback) and sent as Authorization: Bearer <token> on every request.

Run

With uv:

uv run server.py

Or register it with an MCP client (e.g. Claude Code / Claude Desktop) in your MCP config:

{
  "mcpServers": {
    "agent-dashboard": {
      "command": "uv",
      "args": ["--directory", "/path/to/agent-dashboard-mcp", "run", "server.py"],
      "env": {
        "DASHBOARD_API_URL": "http://localhost:3420",
        "DASHBOARD_TOKEN_PATH": "/path/to/.dashboard-token"
      }
    }
  }
}

Expected API shape

The server targets a REST API with these routes (adapt the wrapper if yours differs):

| Tool group | Method + path | |------------|---------------| | memory | GET/POST /api/memories | | kanban | GET/POST /api/kanban, PUT /api/kanban/:id, POST /api/kanban/:id/move\|archive\|comments | | messages | GET/POST /api/messages, PUT /api/messages/:id | | daily log | POST /api/daily-log | | agents | GET /api/agents |

bot_tmux_status and bot_log_tail run locally (tmux / filesystem), not over HTTP.

License

MIT — see LICENSE.

MCP Servers