A collaborative MCP server for sharing AI agent contexts between engineers

## 📋 Motivation & Overview Modern programming tools like Cursor, Windsurf, Augment Code, and Cline all feature powerful "agent modes" that can automatically complete complex tasks based on natural language instructions: - Updating package dependencies and addressing CVE vulnerabilities - Performing routine code refactoring across multiple files - Adding unit tests for new or existing functionality - Standardizing code formatting and fixing linting issues However, these capabilities face a critical limitation: **context sharing**. ### The Problem - Each team member uses different tools (Cursor, Cline, etc.) - Natural language "task templates" exist only in individual environments - No centralized way to share, version, or collaboratively improve these contexts - Knowledge silos form as team members develop their own agent instructions ### The Solution: Egent Egent converts GitHub repository-based contexts into an MCP Server that all mainstream programming tools support. This enables: - **Collaborative editing** of agent contexts through standard GitHub workflows - **Version control** for your team's AI prompts and task templates - **Immediate sharing** of new capabilities across the entire team For example, when engineer Alice adds a new task template for "Updating dependency versions across the monorepo," commits it to the context repo, and pushes, engineer Bob immediately gains access to this capability in his preferred coding tool. ## 🔍 Design

Your team can continuously enrich the knowledge base by editing the context-repo (typically a GitHub repository), enabling AI to automatically complete various tasks. For context repository structure details, see [docs/context.md](docs/context.md). ## 🚀 How to Use ### Setup #### Basic Configuration To use Egent with a remote context repository: ```json { "mcpServers": { "egent": { "command": "npx", "args": ["-y", "egent@latest", "--context-repo", ""] } } } ``` #### Local Development For testing with a local context directory: ```json { "mcpServers": { "egent-local": { "command": "npx", "args": [ "-y", "egent@latest", "--context-path", "" ] } } } ``` #### Egent Development For developing Egent itself: ```json { "mcpServers": { "egent-dev": { "command": "node", "args": ["build/index.js", "--context-path", ""] } } } ``` You can also use `inspector` to inspect the MCP resources: ```bash npx @modelcontextprotocol/inspector node build/index.js --context-repo git@github.com:xuezhaojun/egent-context.git ``` ### Chat with Your Code-Agent Below is an example configuration in Cursor: ```json { "mcpServers": { "egent": { "command": "npx", "args": [ "-y", "egent@latest", "--context-repo", "git@github.com:xuezhaojun/egent-context.git" ] } } } ``` The first command must be `egent_start` to initiate interaction: ``` egent_start Say Hi to Egent. ``` Or more specific: ``` use MCP tool egent_start Say hi to Egent. ``` Your code-agent will then add a comment on this [issue](https://github.com:xuezhaojun/egent-context/issues/2).

For all tools supported by Egent, see the [MCP documentation](docs/mcp.md). ## 💡 Recommended Practices - **Build a task library**: Create templates for common operations like setting up new components, implementing authentication flows, or generating test suites - **Share team knowledge**: Document project-specific patterns and practices as context for the agents - **Cross-tool compatibility**: Ensure all team members benefit regardless of their preferred coding tool - **Iterative improvement**: Continuously refine task templates based on team feedback

Context Sharing

Egent