code-intel-mcp

Code Intelligence MCP Server — Joern CPG + ArangoDB powered code analysis for AI agents.

14 tools for deep code understanding: symbol search, call graphs, data flow tracking, impact analysis, React component trees, and more. Designed for use with Claude Code, Cursor, OpenCode, or any MCP-compatible AI agent.

Why code-intel-mcp?

Capability	grep/ripgrep	AST tools	code-intel-mcp
Find symbol by name	partial	exact	exact + fuzzy
Call graph (who calls X?)	no	single file	multi-file, transitive
Data flow (where does this var go?)	no	no	yes — taint tracking
Impact analysis (what breaks if X changes?)	no	no	yes — transitive blast radius
React component tree	no	partial	JSX-aware
Cross-file call chain A→B	no	no	pathfinding
Incremental re-indexing	N/A	N/A	SHA256 diff — only changed files

Quick Start

Option 1: Docker (recommended)

Everything in one container — Joern, ArangoDB, and the MCP server:

docker compose up -d

Then add to your MCP client config:

{
  "mcpServers": {
    "code-intel": {
      "command": "docker",
      "args": ["exec", "-i", "code-intel-mcp", "node", "dist/index.js"],
      "env": {}
    }
  }
}

Environment Variables

Variable	Default	Description
JOERN_CLI_PATH	/opt/joern/joern-cli	Path to Joern CLI installation
ARANGO_HOST	http://localhost:8529	ArangoDB host URL
ARANGO_USER	root	ArangoDB username
ARANGO_PASS	(empty)	ArangoDB password
ARANGO_DB	code_intel	ArangoDB database name

Copy .env.example to .env and fill in your values.

Tools Reference

Indexing

Tool	Use When	Don't Use When
index_project	First time setup or after code changes	No files changed (check project_status first)
project_status	Before any query to verify index exists	You just indexed and know it's current
list_files	Verify what's indexed or explore file structure	Searching for a specific symbol (use symbol_search)
cache_stats	Debugging query performance	Routine use

Search & Discovery

Tool	Use When	Don't Use When
symbol_search	You know or partially know a symbol name	You already have the exact name from prior search
find_usages	"Where is X used?" — calls, imports, references	You need call graph depth traversal (use get_callers)
get_code_context	Starting a new task — orient before diving in	You already know which files matter

Call Graph

Tool	Use When	Don't Use When
get_callers	"Who calls X?" with depth traversal needed	Simple "where is X used?" (use find_usages — faster)
get_callees	"What does X call?" with depth traversal	React component children (use get_react_components)
get_call_chain	"How does A reach B?" — specific path	Open-ended exploration (use get_callers depth>1)

Analysis

Tool	Use When	Don't Use When
get_data_flow	Security/taint tracking, variable propagation	Simple "where is X set?" (use find_usages)
get_impact_analysis	Before modifying shared code — blast radius	Single-file changes or you already know dependents
get_react_components	React/Next.js component tree exploration	Non-React projects (use get_callees instead)
get_hook_usage	React hook adoption patterns (who uses useAuth?)	Non-hook searches (use symbol_search)

get_data_flow params: { sourceName: string, functionName: string, projectPath: string, direction?: "forward" | "backward" | "both" }

• sourceName — variable or parameter to trace
• functionName — function containing that variable (required)
• direction — forward (where data goes), backward (where it comes from), both (default: forward)

Quick Refer

... View full README on GitHub