Mesh — AI-Native Context Protocol (MCP)

Overview

Mesh MCP is a specialized distribution of the Mesh Agent that bridges its advanced engineering capabilities directly into your favorite AI-native IDEs and CLIs.

It doesn't just provide files; it provides a structured engineering context layer. It includes over 112 specialized tools for codebase mapping, surgical patching, speculative timelines, autonomous discovery, and runtime debugging.

Quick Start

# Install Mesh into all detected MCP clients (Claude Desktop, Cursor, etc.)
npx @edgarelmo/mesh-mcp install

# Run diagnostics to ensure your environment is ready
npx @edgarelmo/mesh-mcp doctor

Core Capabilities (Now via MCP)

By running Mesh as an MCP server, your AI assistant gains immediate access to the entire toolset of the Mesh CLI:

🧠 Workspace Intelligence

• Semantic Mapping: Hybrid ranked file search, symbol dependency chains, and digital twin building.
• AST Compression: High-performance code compression that fits large codebases into small token windows.
• Causal Intelligence: Understand not just what the code is, but why it is structured that way.

⚡️ Safe Code Engineering

• Surgical Patching: Use workspace.patch_surgical for pinpoint accurate edits without token waste.
• Isolated Timelines: Test changes in speculative realities before promoting them to your main branch.
• Reality Forks: Explore multiple alternative implementation paths for a single intent.

🕵️ Runtime & Debugging

• Hologram Tracing: Start commands under observation to capture stack traces and causal autopsies.
• Predictive Repair: Surface and fix failing diagnostics before they become bugs.
• Self-Defending Code: Proactively probe and harden security-sensitive patterns.

Prompts (Slash Commands)

Mesh MCP exports high-level engineering workflows as MCP Prompts. In clients like Claude Code, these appear as guided actions:

| Prompt | CLI Equivalent | Description | |--------|----------------|-------------| | mesh_debug_issue | /fix | Systematic debugging with progressive context. | | mesh_implement_change | /implement | Implementation with targeted reads and impact analysis. | | mesh_review_diff | /review | Deep review of uncommitted changes and risk assessment. | | mesh_ask_codebase | /ask | Semantic query over the entire codebase intelligence. | | mesh_synthesize | /synthesize | Auto-generate full-stack updates from raw intent. | | mesh_setup | /setup | Initialize and index the current workspace. |

Client Configuration

Claude Desktop / Claude Code

The npx @edgarelmo/mesh-mcp install command handles this automatically. For manual setup:

{
  "mcpServers": {
    "mesh": {
      "command": "npx",
      "args": ["-y", "@edgarelmo/mesh-mcp"],
      "env": {
        "MESH_WORKSPACE_PATH": "/absolute/path/to/your/project"
      }
    }
  }
}

Cursor / Windsurf

Add a new MCP server in settings:

• Type: stdio
• Command: npx -y @edgarelmo/mesh-mcp
• Env: MESH_WORKSPACE_PATH=/your/project/path

Requirements

• Node.js 20 or newer.
• For local development and the latest features, Mesh MCP automatically bridges to your local @edgarelmo/mesh-agent-cli if found on your desktop.

License

UNLICENSED. All rights reserved.

This repository and npm package are not released under the MIT license or any other open-source license.