Code intelligence for AI agents

64 MCP tools that give Claude Code, Cursor, Codex, and Gemini deep understanding of your codebase — with 61% fewer tokens than native workflows.

No signup. Local-first. Install in 30 seconds.
$ npm install -g codesift-mcp
View on GitHub
terminal
Works with
Claude Code Cursor Codex

AI agents are flying blind

Your coding agent doesn't understand your codebase structure. It uses grep to search — missing semantic meaning. It reads whole files — wasting tokens. It can't trace call chains, detect secrets, search past conversations, or understand module boundaries. CodeSift gives it 64 tools that change how it navigates code.

What you get

64 tools, one MCP server

Intelligent Search

requires setup

BM25F ranking with centrality bonus. Semantic search with embeddings. Token-budget-aware results that cut 59% of tokens.

Code Graph

Call chains, impact analysis, HTTP route tracing. Trace any endpoint from handler to database in one call.

LSP Bridge

requires setup

Go-to-definition, type info, cross-file rename. Language server precision for 6 languages, zero setup.

Analysis Suite

Dead code, complexity ranking, clone detection, hotspots, AST queries, and 9 built-in anti-patterns.

Security Scanning

~1,100 secret detection rules with AST-aware confidence scoring. Masked output. Never logs raw values.

Conversation Memory

Search past AI coding sessions. Cross-reference code symbols with the discussions that shaped them.

Cross-Repo Search

Search symbols and text across all indexed repos. One query, all your projects.

Context Levels

L0 full source to L3 directory overview. Control token density per query. 3x more symbols at L1.

Benchmarked

Numbers, not promises

-59%
Fewer tokens
vs native workflows (combo benchmark)
64
MCP tools
search, graph, analysis, LSP
5.6x
Faster reindex
incremental mtime-based
7.8/10
Search quality
vs 6.5/10 baseline
See full results

Get started

Three steps. Thirty seconds.

1

Install

$ npm install -g codesift-mcp
2

Add to your MCP config

{
  "mcpServers": {
    "codesift": {
      "command": "codesift-mcp"
    }
  }
}
3

Ask your AI agent anything

// Ask your AI agent:
"Search for authentication logic in the codebase"

// CodeSift returns ranked results:
// ✓ src/auth/guard.ts — AuthGuard (BM25: 0.92)
// ✓ src/middleware/session.ts — validateSession (BM25: 0.87)
// ✓ src/utils/jwt.ts — verifyToken (BM25: 0.81)
// Total: 1,291 tokens (vs 2,985 with grep)

Compare

CodeSift vs traditional tools

Capability CodeSift grep / Read
Semantic search
AST-aware parsing
Token-budget control
Call chain tracing
Cross-repo search
LSP integration
Dead code detection
Clone detection
Complexity ranking
Secret scanning (~1,100 rules)
Conversation memory
Architecture boundary checks

Pricing

Free to start. Pro when you're ready.

Free

$0

Free for personal use

  • 64 MCP tools
  • 3 repos indexed
  • BM25F search with centrality bonus
  • File outlines and symbol retrieval
  • Call chain tracing
  • Impact analysis
npm install -g codesift-mcp
Coming soon

Pro

$79/yr

For professional developers

  • Everything in Free
  • Unlimited repos
  • Semantic search (embeddings)
  • Full analysis suite (dead code, clones, hotspots)
  • LSP bridge (go-to-def, rename, type info)
  • All Zuvo audit skills
  • Cross-repo search
  • Priority support