Octocode includes a built-in MCP server that exposes AI-assisted code understanding through the Model Context Protocol. To enable the MCP capabilities, start the server using the provided binary (for Rust builds this is typically the octocode-mcp binary). Once running, you can interact with the MCP endpoints to trigger AI-assisted code analysis, memory-enabled semantic search, and LSP-based code intelligence within your editor or IDE. The MCP integration is designed to work alongside Octocode’s index/search features, enabling context-aware code completion, smart assistance during reviews, and conversational queries about your codebase.

Typical workflow:

• Start the MCP server to boot the LSP and AI services for your project.
• Use your editor’s MCP can be configured to communicate with the running MCP endpoint to fetch context-aware suggestions, code snippets, and relationship graphs derived from your repository.
• Leverage the memory system and graph-based reasoning to surface connections between files, modules, and functions, improving navigation and understanding of large codebases.

Prerequisites:

• Rust toolchain (rustc and cargo) installed on your system
• git installed
• Optional: Voyage API key for embeddings and optional OpenRouter API key for enhanced AI features

Installation steps:

1. Clone the repository or download the source: git clone https://github.com/Muvon/octocode.git cd octocode
2. Build the MCP server (Rust/Cargo-based): cargo build --release

   or install the MCP binary if provided by the project:

   cargo install --path . --bin octocode-mcp

3. Ensure dependencies and environment are set:
   • Set required API keys if you plan to use embeddings or AI features: export VOYAGE_API_KEY="your-voyage-api-key" export OPENROUTER_API_KEY="your-openrouter-api-key" (optional)
4. Run the MCP server (example):

   From the project root, using the built binary

   cargo run --bin octocode-mcp -- --path /path/to/your/project

   Or, if installed:

   octocode-mcp --path /path/to/your/project
5. Verify the server is listening on the configured port and connect your editor/clients using the MCP protocol endpoints defined by the project documentation.

Notes and tips:

• The MCP server is designed to run locally with a local-first design, and will not access external networks for search unless you configure embedding providers like Voyage or OpenRouter.
• Ensure your VOYAGE_API_KEY is set if you plan to use embedding-based features; tokens are typically metered by the provider.
• If you encounter port conflicts, change the MCP server binding port in the configuration or environment variables as documented in the MCP integration guide.
• The server exposes LSP capabilities for language-aware assistance; configure your editor's LSP client to point to the MCP server endpoint for real-time code intelligence.
• Check logs for graph construction and memory system operations to troubleshoot index or semantic search behavior.