Updated November 18, 2026
Glossary ToolsBest models

Binary Ninja

Binary Ninja AI with BinAssist, MCP and Sidekick

For AI in Binary Ninja, use BinAssist inside the app or Binary Ninja MCP so Claude Code can drive it. Sidekick uses a Sidekick account and its own models.

By AI for Reverse EngineeringUpdated November 18, 2026
On this page

BinAssist, an LLM panel

BinAssist supports "any OpenAI v1-compatible or Anthropic API." It needs Binary Ninja 4000 or later.

  1. Install it, then install its dependencies from the plugin folder: pip install -r requirements.txt.
  2. Open SettingsBinAssist and set the provider:
    FieldValue
    API Hosthttps://api.silkcompute.com (or https://api.silkcompute.com/v1 if your version asks for the path)
    API KeyYour key
    Modelkimi-k3.1-pro-uncensored
  3. Set the database paths for its RAG and feedback features, and adjust token limits.
  4. Load a binary and open BinAssist from its sidebar icon.

jtang613/BinAssist README, read November 18, 2026.

Binary Ninja MCP

Binary Ninja MCP exposes HLIL, MLIL and LLIL, plus imports and exports. Its README links a CTF demo.

  1. Install your MCP client first, then the plugin from PluginsManage Plugins. It sets up supported clients, Claude Code among them, on its own.
  2. Or add it by hand. The README's config runs npx -y binary-ninja-mcp --host localhost --port 9009; in Claude Code:
    terminalshell
    claude mcp add binja -- \
      npx -y binary-ninja-mcp --host localhost --port 9009
  3. Point Claude Code at an abliterated model with the settings on the MCP setup page.

fosdickio/binary_ninja_mcp README, read November 18, 2026.

Sidekick

Sidekick is Vector 35's paid extension. Its settings reference lists no field for your own model endpoint, read November 18, 2026. A local model works in BinAssist or Binary Ninja MCP.