agent-delegate
agent-delegate is an MCP-first delegation advisor for coding agents. It helps a main agent decide when to delegate read-only investigation, generate bounded subagent briefs, and summarize subagent results back into compact main-agent context.
README
agent-delegate
agent-delegate is an MCP-first delegation advisor for coding agents. It helps a main agent decide when to delegate read-only investigation, generate bounded subagent briefs, and summarize subagent results back into compact main-agent context.
Status
This repository implements v1: Read-only Delegation Advisor.
v1 does not execute write-code subagents, edit files on behalf of agents, run background monitoring, or orchestrate worktrees.
Install
npm install
Run the MCP Server
npm run --silent agent-delegate -- serve
Configure your agent tool to launch that command as a local MCP server.
Check MCP Setup
npm run agent-delegate -- doctor
doctor verifies that dependencies are installed, the local MCP server can start, and all five v1 MCP tools are visible through stdio. It also prints a copyable MCP configuration.
Analyze a Sample Agent State
npm run agent-delegate -- analyze examples/agent-state.sample.json
Expected output includes a recommendation such as:
{
"recommendation": "dispatch_readonly"
}
Dogfood Before Release
This project should be dogfooded before any open-source release. Unit tests are not enough because the core product is an MCP tool used by real agents.
Run the MCP smoke test:
npm run smoke:mcp
Run realistic fixtures:
npm run check:fixtures
See docs/dogfood.md for the full dogfood protocol.
Generate Rules
npm run agent-delegate -- init codex
npm run agent-delegate -- init claude-code
npm run agent-delegate -- init generic
MCP Tools
record_eventassess_delegation_needgenerate_delegation_briefsassess_brief_qualitysummarize_subagent_results
Documentation
- Dogfood protocol
- Dogfood results
- MCP configuration
- Release checklist
- Contributing
- Security policy
- Changelog
Development
npm test
npm run typecheck
npm run build
Roadmap
- v1: Read-only Delegation Advisor
- v2: Write-readiness Assessor
- v3: Controlled Write-agent Delegation
- v4: Parallel Write-agent Orchestration
推荐服务器
Baidu Map
百度地图核心API现已全面兼容MCP协议,是国内首家兼容MCP协议的地图服务商。
Playwright MCP Server
一个模型上下文协议服务器,它使大型语言模型能够通过结构化的可访问性快照与网页进行交互,而无需视觉模型或屏幕截图。
Magic Component Platform (MCP)
一个由人工智能驱动的工具,可以从自然语言描述生成现代化的用户界面组件,并与流行的集成开发环境(IDE)集成,从而简化用户界面开发流程。
Audiense Insights MCP Server
通过模型上下文协议启用与 Audiense Insights 账户的交互,从而促进营销洞察和受众数据的提取和分析,包括人口统计信息、行为和影响者互动。
VeyraX
一个单一的 MCP 工具,连接你所有喜爱的工具:Gmail、日历以及其他 40 多个工具。
graphlit-mcp-server
模型上下文协议 (MCP) 服务器实现了 MCP 客户端与 Graphlit 服务之间的集成。 除了网络爬取之外,还可以将任何内容(从 Slack 到 Gmail 再到播客订阅源)导入到 Graphlit 项目中,然后从 MCP 客户端检索相关内容。
Kagi MCP Server
一个 MCP 服务器,集成了 Kagi 搜索功能和 Claude AI,使 Claude 能够在回答需要最新信息的问题时执行实时网络搜索。
e2b-mcp-server
使用 MCP 通过 e2b 运行代码。
Neon MCP Server
用于与 Neon 管理 API 和数据库交互的 MCP 服务器
Exa MCP Server
模型上下文协议(MCP)服务器允许像 Claude 这样的 AI 助手使用 Exa AI 搜索 API 进行网络搜索。这种设置允许 AI 模型以安全和受控的方式获取实时的网络信息。