code-context-mcp

MCP server for semantic code search and dependency graph analysis. Indexes codebases into a knowledge graph (Apache AGE) with vector embeddings (pgvector + Ollama) for AI-powered code understanding.

Supported Languages

• TypeScript (.ts, .tsx) — functions, classes, types, interfaces, imports
• Rust (.rs) — functions, structs, enums, traits, impl blocks, modules, use declarations
• Solidity (.sol) — contracts, functions, events, modifiers, structs, enums, interfaces, imports
• C - functions, enums

Prerequisites

• Docker and Docker Compose
• Node.js >= 18

Quick Start

# Start PostgreSQL (with AGE + pgvector) and Ollama
docker-compose up -d

# Install dependencies and build
npm install
npm run build

Indexing a Codebase

# Index a directory (project name auto-derived from directory basename)
node dist/cli/index.js index ./src

# Override the project name
node dist/cli/index.js index ./src --project my-project

# Index multiple codebases
node dist/cli/index.js index /path/to/frontend --project frontend
node dist/cli/index.js index /path/to/backend --project backend

The indexer will:

1. Parse all supported source files (TypeScript, Rust, Solidity) using tree-sitter
2. Build a code graph with functions, classes, types, contracts, traits, and their relationships
3. Generate vector embeddings for semantic search via Ollama
4. Show a progress bar during indexing

Re-running the indexer on the same directory skips unchanged files (detected via content hash).

Claude Desktop Configuration

Add to your Claude Desktop config (~/Library/Application Support/Claude/claude_desktop_config.json):

{
  "mcpServers": {
    "code-context": {
      "command": "node",
      "args": ["dist/mcp/server.js"],
      "cwd": "/path/to/code-context-mcp",
      "env": {
        "PGHOST": "localhost",
        "PGPORT": "5433",
        "PGDATABASE": "rag_db",
        "PGUSER": "postgres",
        "PGPASSWORD": "postgres",
        "OLLAMA_URL": "http://localhost:11434"
      }
    }
  }
}

MCP Tools

list_projects

List all indexed projects and their node counts.

Parameters: none

search_code

Semantic search across the indexed codebase. Finds functions, classes, types, and imports by meaning.

Parameter	Type	Required	Description
query	string	yes	Natural language search query
limit	number	no	Maximum results (default: 10)
project	string	no	Scope search to a project

get_call_stack

Trace the call dependency tree from a function.

Parameter	Type	Required	Description
functionName	string	yes	Function to trace from
depth	number	no	Max traversal depth, 1-10 (default: 3)
project	string	no	Scope to a project

get_function_context

Gather full context for a function: callers, callees, types it uses.

Parameter	Type	Required	Description
functionName	string	yes	Function to get context for
project	string	no	Scope to a project

get_impact_analysis

Analyze impact of changes to a file. Shows all external code that depends on definitions in the file.

Parameter	Type	Required	Description
filePath	string	yes	Relative path of the file (as indexed)
project	string	no	Scope to a project

Environment Variables

Variable	Default	Description
PGHOST	localhost	PostgreSQL host
PGPORT	5433	PostgreSQL port
PGDATABASE	rag_db	Database name
PGUSER	postgres	Database user
PGPASSWORD	postgres	Database password
OLLAMA_URL	http://localhost:11434	Ollama API endpoint

Testing

Unit Tests

Unit tests run without Docker and cover extractors, scanner, registry, progress bar, embeddings, and DB queries.

npm test              # Run all unit tests
npm run test:watch    # Watch mode

Integration Tests

Integration tests require separate Docker containers to avoid interfering with development data.

# Start test infrastructure (separate ports: PG=5434, Ollama=11435)
docker compose -f docker-compose.test.yml up -d

# Run integration tests
npm run test:integration

Integration tests cover the full indexing pipeline and multi-language graph population.