Code Intelligence for AI Agents
Deep static analysis, security scanning, and semantic search — all through the Model Context Protocol
$ brew tap postrv/narsil && brew install narsil-mcp
$ narsil-mcp --repos ~/your-projectAI coding assistants are flying blind
Without deep code intelligence, AI assistants struggle with complex codebases
AI reads files one at a time
AI queries semantic index instantly
Security issues discovered late
Continuous vulnerability scanning
"Find where X is defined" takes 30 seconds
Symbol lookup in <1ms
No call graph awareness
Full call graph with transitive analysis
Manual dependency audits
Automated SBOM + CVE detection
90 tools. One MCP server.
Everything your AI assistant needs to understand your codebase
Semantic Search
BM25 + TF-IDF + neural embeddings. Find code by meaning, not just text.
Security Scanning
OWASP Top 10, CWE Top 25, secrets detection, taint analysis.
Call Graphs
Who calls what? Trace paths between functions. Find hotspots.
Supply Chain
SBOM generation, OSV vulnerability checks, license compliance.
Type Inference
Infer types in Python/JS/TS without running external checkers.
Git Integration
Blame, history, hotspots, contributors — all queryable.
32 Languages
Rust, Python, TypeScript, Go, Java, C#, Ruby, Kotlin, Swift, and 23 more.
100% Local
Your code never leaves your machine. No cloud required.
From install to insights in 60 seconds
Get started with narsil-mcp in three simple steps
Install
$ brew tap postrv/narsil
$ brew install narsil-mcp
# Or cargo, npm, scoop, nix...Configure
# claude_desktop_config.json
{
"mcpServers": {
"narsil": {
"command": "narsil-mcp",
"args": ["--repos", "~/project"]
}
}
}Ask Claude
"What security vulnerabilities
exist in the auth module?"
→ 3 findings: CWE-89,
CWE-79, CWE-352The CCG Standard
Publish your codebase for AI
The Code Context Graph (CCG) is an open standard for publishing machine-readable code intelligence.
- Symbols, call graphs, security findings — all in one file
- Works with any LLM, any MCP client
- Auto-generated by narsil-mcp, hosted on codecontextgraph.com
{
"@context": "https://codecontextgraph.com",
"type": "CodeRepository",
"name": "my-project",
"symbols": [...],
"callGraph": {...},
"security": {
"findings": [...]
}
}How narsil-mcp compares
Feature comparison with other MCP code intelligence servers
| Feature | narsil-mcp | GitHub MCP | XRAY | Serena |
|---|---|---|---|---|
| Languages | 32 | N/A | 4 | 30+ |
| Neural search | ||||
| Taint analysis | ||||
| SBOM/Licenses | Partial | |||
| Offline/Local | ||||
| WASM/Browser | ||||
| Call graphs | Partial | |||
| Type inference |
Data as of January 2026. Contribute corrections via GitHub