Sugar MCP
A shared task board and memory for agents: Claude Code, Codex, Cursor and Goose take work from one queue with history and costs
- 101 ★ on the repository
- Last commit
Broad access
Why this level
- `sugar run` autonomously takes tasks, edits code, runs tests and makes commits without stopping
- The log and memory hold project context and decisions and should be protected as working data
Install
- Works in
In your terminal, with SkillFoxx CLI
npx skillfoxx add mcp/sugarDetects the agents on your machine, shows the access level and pins the version.
Other ways to install
Access and checks
How we reviewBroad accessThe tool can change data, run commands or handle secrets.
- `sugar run` autonomously takes tasks, edits code, runs tests and makes commits without stopping
- The log and memory hold project context and decisions and should be protected as working data
This is third-party code. Review the repository files before installing.
What it does
Sugar puts a persistent SQLite task queue between agents and the repository. An agent takes a task atomically with a lease and a heartbeat: if the worker dies, the lease expires and the task returns to the queue. The same actions are available from the CLI (`sugar add`, `sugar list`, `sugar run`), through two MCP servers for tasks and memory, over REST and SSE and in a web dashboard with a live feed of agent actions. Every status change and harness event goes to an append-only log: who, with which harness, on which task and at what cost. Costs come from the agent own reports with no invented pricing, and `sugar audit export` writes tamper-evident JSONL with a SHA-256 manifest. Memory is split into project and global, seven record types, searched by local embeddings with an FTS5 fallback. The agent reads and writes it through `search_memory`, `store_learning` and `recall`.
Who it is for. People running several agents on one repository who want to see who took what, how it ended and what it cost.
Good fit when
- Two agents pick up the same task and overwrite each other
- A crashed run keeps a task claimed forever and the queue stalls
- A history is needed to answer who did what yesterday without scrolling terminals
- The agent starts every session from scratch and last week decisions are lost
- Costs by agent and project must be exported with a signature for a report
Not a fit when
- You work with one agent in one session: a queue and leases are redundant
- A cloud service with accounts is needed: Sugar is local and the team edition is self-hosted
- The AGPL-3.0 license does not fit a closed product: a separate commercial one exists for it
Example request
Take the next high-priority task from the Sugar queue, complete it and store in memory the decision about handling the authorization timeoutMCP
- Transport
- stdio
- Authentication
- not required
Limitations
Python 3.11 or newer is required, and the app needs macOS 14. The license is dual: AGPL-3.0 and commercial. `sugar run` currently drives only Claude Code through the Agent SDK, while other harnesses work on their own and send events through `sugar emit`. Autonomous run takes tasks, executes, tests and commits without stopping, and `--ralph` repeats the loop until done. Semantic memory search needs the `sugarai[memory]` extra. Cost tracking relies on what the harness itself reports.
How to disable. Remove the servers with `claude mcp remove sugar` and `claude mcp remove sugar-tasks`, then `pipx uninstall sugarai`; data lives in the project `.sugar` and `~/.sugar`.
FAQ
How do I connect Sugar memory and tasks to Claude Code?
With two commands: `claude mcp add sugar -- sugar mcp memory` and `claude mcp add sugar-tasks -- sugar mcp tasks`, after installing the package with `pipx install 'sugarai[all]'`.
Which agents work with the queue?
Claude Code, Codex, Cursor, Goose, OpenCode and any MCP client. OpenCode has `sugar opencode setup`.
What does `sugar run` do?
It starts an autonomous runner that takes a task, executes it through Claude Code, tests, commits and moves to the next one.
Is an account needed?
No, Sugar is local: data stays in your files and no account is needed.
What license applies?
Dual: AGPL-3.0 and a commercial one for closed use.
Discussion
No questions or reviews yet.
Related
MCP serversA memory and coordination layer for coding agents: specs and issues live in git while agents work in parallel
Long-term memory for agents: new facts supersede old ones, every answer traces to its source, and recall makes no LLM calls
Shared memory for coding agents: messages in SQLite, old context compacted into a summary tree and MCP tools to search it.
MCP server with a memory of the agent's own: it decides what to keep and wakes with it each session, stored in one SQLite file
Long-term memory for agents: it distills knowledge from session history and brings it back in new tasks, one store shared across Claude Code, Codex, Cursor and
External memory for Hermes, Claude Code, Codex and Cursor: session archive, recall layers and relevant context injected back
README in short
The problem of agents that cannot see each other and the Sugar design in five layers: queue, dashboard, history, connection and memory. Quick start, the live fleet view with `sugar emit`, a self-hosted team edition, requirements and license.