orchestrator-mcp
Multi-model agent orchestration MCP server that enables plan-code-review-deliver pipelines with configurable providers and models.
README
orchestrator-mcp
多模型 Agent 编排 MCP:plan → code → review → deliver。
- Handoff schema 固定(
schemas/) - Provider 按厂商(deepseek / moonshot / zhipu / openai),model 按 stage 配置
- 别名:
glm→ zhipu,gpt→ openai - 凭证:环境变量 → WebUI 本地 JSON →
~/Desktop/服务器.md - Web 配置界面:编辑 Provider Key / Base URL / 默认模型,以及各 Profile 的 Stage 模型
快速开始
cd orchestrator-mcp
./verify.sh # 离线 stub 自测(含 WebUI API 冒烟)
./start.sh # MCP :18067
./start-webui.sh # 配置 WebUI :18068 → http://127.0.0.1:18068
# 可选:只测 plan 阶段真实 DeepSeek 调用
ORCHESTRATOR_LIVE_TEST=1 ./verify.sh
Web 配置界面
| 功能 | 说明 |
|---|---|
| Providers | 编辑 api_key、base_url、default_model;写入 data/providers.local.json |
| Stages | 按 Profile 覆盖 plan/code/review/deliver 的 provider + model;写入 data/stages.local.json |
本地配置文件不进 git。环境变量仍优先于 WebUI 写入的值。
Profile
| Profile | 用途 |
|---|---|
daily-dev-stub |
离线自测,不调 API |
daily-dev |
deepseek plan / glm code / gpt review / moonshot deliver |
example-kimi-plan |
示例:plan 换 kimi |
同一 OpenAI 账号下 plan 用 gpt-4o-mini、code 用 gpt-5:只改 YAML 里各 stage 的 model,不用建两个 provider。
stages:
plan:
provider: gpt
model: gpt-4o-mini
code:
provider: gpt
model: gpt-5
凭证环境变量
| Provider | Env | 默认 Base URL |
|---|---|---|
| deepseek | DEEPSEEK_API_KEY |
https://api.deepseek.com/v1 |
| moonshot | MOONSHOT_API_KEY |
https://api.moonshot.cn/v1 |
| zhipu/glm | ZHIPU_API_KEY |
https://open.bigmodel.cn/api/paas/v4 |
| openai/gpt | OPENAI_API_KEY |
https://api.openai.com/v1 · wire_api=chat |
| codex-lb/codex | CODEX_LB_API_KEY |
https://codex-lb.vvicat.dev/backend-api/codex · wire_api=responses |
Codex 中转(Responses API)
与 Codex CLI 配置对应关系:
model = "gpt-5.4"
model_reasoning_effort = "medium"
model_provider = "codex-lb"
env_key = "CODEX_LB_API_KEY"
[model_providers.codex-lb]
base_url = "https://codex-lb.vvicat.dev/backend-api/codex"
wire_api = "responses"
在 WebUI Providers → Codex 中转 填写:
| 字段 | 值 |
|---|---|
| API 密钥 | 同 CODEX_LB_API_KEY |
| Base URL | https://codex-lb.vvicat.dev/backend-api/codex(不要写成 /code) |
| 默认模型 | gpt-5.4 |
| Wire API | responses |
| Reasoning Effort | medium |
Stage 里把 review/deliver 的 provider 选 codex-lb(或别名 codex),model 填 gpt-5.4。
注意:/chat/completions 在该中转上返回 405;必须用 Responses 协议。
可选覆盖:CODEX_LB_BASE_URL、DEEPSEEK_BASE_URL、MOONSHOT_BASE_URL、ZHIPU_BASE_URL、OPENAI_BASE_URL
服务器.md 标签(下一行或同行):deepseekApiKey、moonshotApiKey、zhipuApiKey、openaiApiKey
Workspace 项目上下文(与 IDE 共用源文件)
MCP 不单独维护记忆文件。传入 workspace(或设置 ORCHESTRATOR_WORKSPACE)后,每个 stage 直接从磁盘读取与 Cursor/Codex 相同的源文件并注入 prompt:
| 路径 | 说明 |
|---|---|
AGENTS.md / CLAUDE.md / agent.md |
项目级 agent 说明 |
.cursor/rules/* |
Cursor 规则 |
.learnings/*.md |
self-improving 沉淀 |
profile skills: |
加载对应 SKILL.md 全文 |
| git | branch、dirty files、diff --stat |
orchestrate_run_start/orchestrate_run_pipeline:workspace、extra_context(仅本次 run)orchestrate_workspace_context:预览将读取哪些文件
MCP 工具
| 工具 | 说明 |
|---|---|
orchestrate_list_providers |
各 provider 是否已配置 key |
orchestrate_provider_check |
检查单个 provider |
orchestrate_workspace_context |
预览 workspace 项目上下文 |
orchestrate_run_start / orchestrate_dispatch / orchestrate_run_pipeline |
编排执行(支持 workspace) |
orchestrate_stage_override |
运行时换 provider/model |
Cursor 配置
本仓库已在 .cursor/mcp.json 写好配置(stdio 模式,不用手动 ./start.sh)。
- 用 Cursor 打开
localopenclaw仓库根目录 - Settings → Tools & MCP,确认 orchestrator-mcp 已启用(绿点)
- 若没有,点 Refresh 或重启 Cursor
若仍看不到:Settings → MCP → Edit Config,确认项目级 .cursor/mcp.json 已加载。
方式 A(推荐,已写入项目):Cursor 自动拉起进程
{
"mcpServers": {
"orchestrator-mcp": {
"command": ".../orchestrator-mcp/.venv/bin/python",
"args": ["-m", "orchestrator_mcp"],
"env": {
"PYTHONPATH": ".../orchestrator-mcp/src",
"ORCHESTRATOR_MCP_TRANSPORT": "stdio"
}
}
}
}
方式 B:先 ./start.sh,再用 URL(需保持终端进程运行)
{
"mcpServers": {
"orchestrator-mcp": {
"url": "http://127.0.0.1:18067/mcp"
}
}
}
Recommended Servers
playwright-mcp
A Model Context Protocol server that enables LLMs to interact with web pages through structured accessibility snapshots without requiring vision models or screenshots.
Magic Component Platform (MCP)
An AI-powered tool that generates modern UI components from natural language descriptions, integrating with popular IDEs to streamline UI development workflow.
Audiense Insights MCP Server
Enables interaction with Audiense Insights accounts via the Model Context Protocol, facilitating the extraction and analysis of marketing insights and audience data including demographics, behavior, and influencer engagement.
VeyraX MCP
Single MCP tool to connect all your favorite tools: Gmail, Calendar and 40 more.
graphlit-mcp-server
The Model Context Protocol (MCP) Server enables integration between MCP clients and the Graphlit service. Ingest anything from Slack to Gmail to podcast feeds, in addition to web crawling, into a Graphlit project - and then retrieve relevant contents from the MCP client.
Kagi MCP Server
An MCP server that integrates Kagi search capabilities with Claude AI, enabling Claude to perform real-time web searches when answering questions that require up-to-date information.
E2B
Using MCP to run code via e2b.
Neon Database
MCP server for interacting with Neon Management API and databases
Exa Search
A Model Context Protocol (MCP) server lets AI assistants like Claude use the Exa AI Search API for web searches. This setup allows AI models to get real-time web information in a safe and controlled way.
Qdrant Server
This repository is an example of how to create a MCP server for Qdrant, a vector search engine.