Model Worker MCP
Click on "Deploy Server".
Wait a few minutes for the server to deploy. Once ready, it will show a "Started" state.
In the chat, type
@followed by the MCP server name and your instructions, e.g., "@Model Worker MCPsubmit a task to refactor the auth module and save the patch to ./output"
That's it! The server will respond to your query, and you can continue using it as needed.
Here is a step-by-step guide with screenshots.
Model Worker MCP
Model Worker MCP 是一个本机、与 provider 和 model 解耦的 MCP server。coordinator(发起和协调任务的 MCP client)通过短调用提交任务;worker daemon(持有队列、session 和执行状态的后台进程)会在 client 断开后继续执行。V1 以 Claude Agent SDK harness 驱动 kimi-k3,公共协议不绑定该实现。
快速开始
需要 Windows、Node.js 22.x,以及可用的 Claude Code runtime。安装依赖并构建:
npm ci
npm run build
npm link让启动 MCP client 的用户环境包含 MODEL_GATEWAY_API_KEY。项目只把这个变量名作为公共 credential 接口;配置文件和命令示例都不保存 credential 值。然后注册推荐的 strict stdio bootstrap:
codex mcp add model-worker -- model-worker-mcp stdio --request-integrity required
claude mcp add --scope user model-worker -- model-worker-mcp stdio --request-integrity required这两条命令会修改对应 client 的用户配置。strict 模式要求 coordinator 在写请求中提供由本地 request-digest 命令计算的摘要,以便 daemon 在产生副作用前拒绝被改写的参数;摘要不是身份认证。执行前应先阅读客户端注册,确认摘要工作流、兼容模式、scope 和现有同名 server;开发测试只使用临时配置目录。
Related MCP server: MCP Background Job Server
更换网关或模型
端点和模型不写死在代码里。daemon 启动时读取 <数据根>\model-registry.json,文件不存在则使用内置默认值(SFKey + kimi-k3)。换中转站或换模型是复制 config/model-registry.example.json 到数据根、改 endpoints 与 upstream_model_id、再 model-worker-mcp daemon restart,不需要重新构建。
凭据引用必须落在 MODEL_GATEWAY_ 环境变量命名空间内,端点必须是 https;文档不合法时 daemon 明确失败而不会退回默认值。细节见运行与维护。
Capability profiles
Capability profile 是 daemon 强制执行的一组工作区和工具权限,不只是 prompt 建议。
text_only:处理文本和已冻结的附件,不接收 workspace。workspace_read:读取授权 workspace;不给模型 shell 或写入工具。patch_proposal:在独立 Git clone 中修改和测试,返回 patch;不会自动改动原工作区。
MCP tools
model_list:列出可用 model binding 和能力声明。model_health:做本地健康检查。V1 production daemon 固定禁用 live probe;opt-in live probe 只存在于独立测试 harness。task_submit:创建 session 并异步提交首个 task。task_continue:在同一 session head 后继续一轮,不能提高权限。task_get:按 cursor 查询状态、事件、结果和 artifact chunk。task_cancel:请求取消 queued 或 active task;对 terminal task 幂等。
继续阅读
This server cannot be deployed
Maintenance
Related MCP Connectors
MCP Server for an Agent Task Marketplace
Hosted MCP server for task-first delegation to remote workstations and workers.
MCP server for building and testing AI agents with multi-model experimentation and insights.
Nifty's MCP server — exposes tasks, projects, messages, and files as tools for AI agents.
Related MCP Servers
- FlicenseAqualityDmaintenanceAn MCP server that enables LLMs to interact with Agent-to-Agent (A2A) protocol compatible agents, allowing for sending messages, tracking tasks, and receiving streaming responses.528-
- AlicenseAqualityDmaintenanceAn MCP server that enables coding agents to execute and manage long-running shell commands asynchronously with capabilities for process monitoring, interaction, and lifecycle management.714 PyPI13MIT
- AlicenseAqualityBmaintenanceAsync MCP server for running long-running AI tasks with real-time progress monitoring, enabling users to start, monitor, and manage complex AI workflows across multiple models.6388 npm5MIT
- AlicenseNot gradedqualityFmaintenanceMCP server that allows AI coding assistants to schedule deferred tasks such as reminders, shell commands, and AI prompts, executing them as an OS daemon.4 npmMIT