code-context-mcp

<

AST Codebase Intelligence, Symbol Extraction, Cyclomatic Complexity & Refactoring Analysis for LLMs & AI Coding Agents.

💡 Why This Exists & Token Efficiency

Traditional AI coding assistants dump entire raw source files into context windows, wasting tens of thousands of tokens on boilerplate, imports, and irrelevant functions. code-context-mcp introduces a claude-mem style 3-layer workflow that provides surgical code context before editing.

Related MCP server: Aplomo MCP Server

🚀 Quick Start

Multi-IDE Quick Start

Claude Code

Add to your ~/.claude.json:

{
  "mcpServers": {
    "code-context-mcp": {
      "command": "python3",
      "args": ["/path/to/code_context_mcp.py"]
    }
  }
}

Cursor

Go to Settings > Features > MCP and add:

• Name: code-context-mcp

• Type: stdio

• Command: python3 /path/to/code_context_mcp.py

Codex

Add to codex.config.toml:

[mcp.servers.code-context-mcp]
command = "python3"
args = ["/path/to/code_context_mcp.py"]

Antigravity

Add to .antigravity/mcp_config.json:

{
  "mcpServers": {
    "code-context-mcp": {
      "command": "python3",
      "args": ["/path/to/code_context_mcp.py"]
    }
  }
}

OpenCode

Add to opencode.json:

{
  "mcpServers": {
    "code-context-mcp": {
      "command": "python3",
      "args": ["/path/to/code_context_mcp.py"]
    }
  }
}

Windsurf

Add to ~/.codeium/windsurf/mcp_config.json:

{
  "mcpServers": {
    "code-context-mcp": {
      "command": "python3",
      "args": ["/path/to/code_context_mcp.py"]
    }
  }
}

🛠️ How It Works

code-context-mcp runs as a standard Model Context Protocol (MCP) server communicating over stdio. It builds an Abstract Syntax Tree (AST) representation of Python source code locally and computes function metrics without sending data externally.

🧬 3-Layer Workflow

Layer 1: Repository Architecture Summary

# Call tool: summarize_repository_architecture
response = summarize_repository_architecture(root_dir=".")
print(response)
# Returns file tree, line counts, extension distribution, and module entry points.

Layer 2: AST Symbol Outline

# Call tool: extract_python_ast_symbols
response = extract_python_ast_symbols(file_path="code_context_mcp.py")
print(response)
# Returns class definitions, method signatures, decorator usages, and docstrings.

Layer 3: Cyclomatic Complexity & Refactoring Insights

# Call tool: analyze_cyclomatic_complexity
response = analyze_cyclomatic_complexity(file_path="code_context_mcp.py", threshold=10)
print(response)
# Identifies high-complexity methods and recommends specific modular refactorings.

🔧 MCP Tools

🏗️ Architecture

+-------------------------------------------------------+
|                   AI Coding Agent                     |
|  (Claude Code / Cursor / Codex / Antigravity / etc.)  |
+---------------------------+---------------------------+
                            | MCP (stdio)
                            v
+-------------------------------------------------------+
|                   code-context-mcp                    |
|                                                       |
|  +--------------------+       +--------------------+  |
|  |  AST Parser Engine |       | Cyclomatic Engine  |  |
|  +--------------------+       +--------------------+  |
|  |  Symbol Extractor  |       | Refactor Context   |  |
|  +--------------------+       +--------------------+  |
+-------------------------------------------------------+

🛡️ Security

code-context-mcp works 100% offline and locally. Zero telemetry, zero external API requests. See SECURITY.md for vulnerability disclosure policy.

📄 License

Distributed under the MIT License. See LICENSE for more details.--26-7928CA?style=for-the-badge" alt="MCP Spec" />

AST Codebase Intelligence, Symbol Extraction, Cyclomatic Complexity & Refactoring Analysis for LLMs & AI Coding Agents.

💡 Why This Exists & Token Efficiency

Traditional AI coding assistants dump entire raw source files into context windows, wasting tens of thousands of tokens on boilerplate, imports, and irrelevant functions. code-context-mcp introduces a claude-mem style 3-layer workflow that provides surgical code context before editing.

⚡ 3-Layer Context Workflow

1. Layer 1: Structural Scan (summarize_repository_architecture) — High-level file structure, line counts, and extensions map.

2. Layer 2: AST Symbol Outline (extract_python_ast_symbols) — Class definitions, method signatures, docstrings, and line ranges without dumping raw code.

3. Layer 3: Targeted Analysis (analyze_cyclomatic_complexity & search_code_todos_and_hotspots) — Cyclomatic complexity scores and TODO/FIXME hotspot discovery to pinpoint refactoring targets.

📊 Token Savings Comparison

🏗️ Architecture Diagram

flowchart TD
    A[AI Coding Agent / LLM] -->|1. Request Repo Summary| B[summarize_repository_architecture]
    A -->|2. Extract AST Symbols| C[extract_python_ast_symbols]
    A -->|3. Analyze Complexity| D[analyze_cyclomatic_complexity]
    A -->|4. Scan Hotspots| E[search_code_todos_and_hotspots]

    subgraph MCP Server Engine: code-context-mcp
        B --> F[AST & Static Code Parser]
        C --> F
        D --> F
        E --> F
    end

    F -->|Compact Structural Context| A

🛠️ Tools Included

1. summarize_repository_architecture: Scans directory paths, generates file structure breakdown by extension, and computes total lines of code.

2. extract_python_ast_symbols: Parses Python files into Abstract Syntax Trees (AST) to extract classes, base classes, methods, docstrings, function signatures, and line markers.

3. analyze_cyclomatic_complexity: Calculates cyclomatic complexity scores for all functions/methods to identify refactoring candidates and branch depth.

4. search_code_todos_and_hotspots: Detects code comments containing TODO, FIXME, HACK, and XXX across the codebase with exact file and line references.

🚀 Quick Start (1-Click & 1-Command Setup)

1. Claude Code CLI

claude mcp add code-context-mcp -- python -m code_context_mcp

2. Cursor (.cursor/mcp.json)

{
  "mcpServers": {
    "code-context-mcp": {
      "command": "python",
      "args": ["-m", "code_context_mcp"]
    }
  }
}

3. OpenAI Codex CLI (~/.codex/config.toml)

[mcp_servers.code-context-mcp]
command = "python"
args = ["-m", "code_context_mcp"]

4. Google Antigravity IDE (mcp_config.json)

{
  "mcpServers": {
    "code-context-mcp": {
      "command": "python",
      "args": ["-m", "code_context_mcp"]
    }
  }
}

5. Windsurf / Cline

{
  "mcpServers": {
    "code-context-mcp": {
      "command": "python",
      "args": ["-m", "code_context_mcp"]
    }
  }
}