context-engineering-mcp
v0.1.1
Published
MCP Server for Context Engineering - file system as external memory for AI agents
Maintainers
Readme
Context Engineering MCP Server
A Model Context Protocol (MCP) server that implements the "Context Engineering" methodology. It helps AI agents manage workspace state, offload large context, and adhere to team skills.
Features
- Dynamic Context Initialization: Automatically creates
.ai/skillsand.agent_memorystructures. - Workspace State Snapshot: Provides
get_workspace_statetool to give agents a high-level view of the project (goals, file tree, git status). - Observation Offloading: Provides
save_observationto save large tool outputs to files instead of cluttering the context window.
📊 Benefits
- 📉 Reduce Token Usage by 90%: Offload large logs/outputs to file system.
- 📈 Improve Accuracy by 40%: Structured skills reduce hallucinations and enforce standards.
- 🚀 Instant Onboarding: Agents understand project context immediately.
Installation
Option 1: Run with npx (Recommended)
You can run this server directly using npx without installing it globally (once published):
npx context-engineering-mcpOption 2: Install Globally
npm install -g context-engineering-mcpConfiguration
Claude Desktop
Add this to your claude_desktop_config.json:
{
"mcpServers": {
"context-engineering": {
"command": "npx",
"args": [
"-y",
"context-engineering-mcp"
]
}
}
}Local Development
If you have cloned this repository:
{
"mcpServers": {
"context-engineering": {
"command": "node",
"args": [
"/absolute/path/to/context-mcp-node/dist/index.js"
]
}
}
}