Codescope

What it does

Codescope is a high-performance code intelligence layer that replaces traditional embedding-based RAG with a knowledge graph. Instead of guessing relevance via vector similarity, it parses code into a structural graph of functions, calls, and type hierarchies, allowing AI agents to traverse code deterministically and with minimal token overhead.

Tools

• impact_analysis: Calculate the transitive blast radius of a change across the codebase.
• find_callers/find_callees: Traverse the 1-hop call graph to find dependencies.
• search: Perform fuzzy, exact, or neighborhood searches across files and types.
• type_hierarchy: Map out inheritance and implementation chains.
• knowledge: Manage project-level memory and Architecture Decision Records (ADRs).
• code_health: Identify hotspots, churn, and coupling issues.
• sandbox_run: Execute Python, Node, or Bash snippets in a secure subprocess.

Installation

Run the following command to install the binary: curl -fsSL https://raw.githubusercontent.com/onur-gokyildiz-bhi/codescope/main/install.sh | bash

Then initialize your project: codescope init

Claude Desktop Configuration:

{
  "mcpServers": {
    "codescope": {
      "command": "codescope",
      "args": ["mcp"]
    }
  }
}

Supported hosts

Confirmed for Claude Code, Cursor, Codex, Zed, and VS Code.