Binary Ninja

Translates Model Context Protocol (MCP) JSON-RPC requests from LLM clients into HTTP GET/POST calls targeting a local Binary Ninja plugin HTTP server on port 9009. Uses FastMCP framework to expose 64 reverse engineering tools as standardized MCP tool definitions, enabling seamless integration between LLM clients (Claude Desktop, Cline, Cursor, etc.) and Binary Ninja's analysis engine without requiring direct Binary Ninja API knowledge from the LLM.

Exposes Binary Ninja's function analysis capabilities through HTTP endpoints that retrieve detailed metadata about functions in loaded binaries, including function names, type signatures, parameter types, return types, and internal control flow information. The BinaryOperations layer queries Binary Ninja's internal function objects and type system to construct structured JSON responses containing function-level analysis without requiring the LLM to understand Binary Ninja's Python API.

Provides a plugin architecture that allows developers to extend the Binary Ninja MCP bridge with custom tools and analysis capabilities. Developers can register new HTTP endpoints in the BinaryNinjaEndpoints class and expose them as MCP tools through the bridge, enabling custom reverse engineering workflows without modifying the core bridge code. The architecture supports adding new tools by implementing simple HTTP endpoint handlers that follow the existing pattern.

Exposes Binary Ninja's cross-reference (xref) tracking system through HTTP endpoints that identify all locations where a function, variable, or memory address is referenced within a binary. Queries Binary Ninja's internal xref graph to return caller/callee relationships, data references, and control flow dependencies, enabling LLMs to understand data flow and function call chains without manual graph traversal.

Enables LLMs to suggest and apply function renames and type annotations to a loaded binary through HTTP POST endpoints that modify Binary Ninja's internal function objects. The BinaryOperations layer validates rename requests and applies changes to the binary's symbol table, allowing LLMs to improve binary readability by assigning meaningful names based on code analysis without requiring manual Binary Ninja UI interaction.

Provides HTTP endpoints to inspect memory contents and data structures at specific addresses in a loaded binary, with type-aware interpretation using Binary Ninja's type system. Queries memory regions, interprets raw bytes according to inferred or user-defined types, and returns structured representations of data structures, enabling LLMs to understand data layout and contents without manual hex dump parsing.

Exposes HTTP endpoints to retrieve disassembled code for functions or address ranges, returning instruction-level details including mnemonics, operands, and metadata. The BinaryOperations layer queries Binary Ninja's IL (Intermediate Language) and disassembly representations to provide both high-level and low-level code views, enabling LLMs to analyze instruction sequences and understand control flow without requiring manual disassembly parsing.

Provides HTTP endpoints to load, unload, and manage multiple binary files within a single Binary Ninja instance, enabling LLMs to switch between binaries or analyze related binaries in a single session. The plugin maintains a registry of loaded binaries and routes requests to the appropriate binary context, allowing complex analysis workflows that involve multiple executable files or libraries.

Provides automated setup tools that configure MCP clients (Claude Desktop, Cline, Cursor, etc.) to connect to the Binary Ninja MCP bridge, including Python environment management, dependency installation, and client-specific configuration file generation. The auto-setup system detects the client type and generates appropriate configuration files (e.g., claude_desktop_config.json for Claude Desktop) with correct bridge paths and environment variables.

Enables LLMs to receive structured analysis results from Binary Ninja and generate natural language insights about binary behavior, vulnerabilities, and functionality. The system provides example prompts and analysis templates that guide LLMs to ask relevant questions about functions, data structures, and control flow, creating a conversational reverse engineering interface where users can ask high-level questions and receive detailed analysis.

Exposes HTTP endpoints to query Binary Ninja's type system, retrieve struct definitions, and manage custom type annotations. The endpoints allow LLMs to inspect type information for functions and data structures, retrieve struct field layouts, and apply custom type definitions to improve code readability. Integrates with Binary Ninja's type inference engine to provide accurate type information for analysis.