deepcontext

DeepContext is an MCP server that provides semantic, symbol-aware search capabilities for large codebases, integrating with Codex, Claude Code, and other agents. It enables agents to locate relevant code chunks by meaning rather than exact text, reduces token usage by returning precise snippets, and speeds up retrieval with pre-indexed data. The server exposes tools to index, query, monitor, and manage your codebases: index_codebase creates a semantic search index, search_codebase retrieves relevant code given natural language or keywords, get_indexing_status shows progress and file counts, and clear_index removes indexed data for a project. To use DeepContext, obtain an API key from Wildcard, configure your MCP client to point at the server (via npx @wildcard-ai/deepcontext@latest), and provide your API key through environment variables. The server is designed to work with TypeScript and Python projects and uses Tree-sitter parsing, semantic symbol extraction, and a hybrid search pipeline (vector similarity plus BM25) to deliver precise results quickly.

Prerequisites:

• Node.js 20+ (for ES module support and performance)
• npm (comes with Node.js)

Installation steps:

1. Clone the MCP repository (or reference the DeepContext MCP package in your environment):

git clone https://github.com/Wildcard-Official/deepcontext-mcp.git
cd deepcontext

2. Install dependencies and build if necessary:

npm install
# If a build step is required by the project, run:
npm run build

3. Run or integrate the MCP server using the recommended MCP client configuration (example shown below). If you are using a hosted client, you can add the server via the client’s MCP configuration command:

claude mcp add deepcontext \
  -e WILDCARD_API_KEY=your-wildcard-api-key \
  -- npx @wildcard-ai/deepcontext@latest

Alternatively, for Codex users, add to ~/.codex/config.toml:

[mcp_servers.deepcontext]
command = "npx"
args = ["-y", "@wildcard-ai/deepcontext@latest"]
env = { "WILDCARD_API_KEY" = "your-wildcard-api-key" }

4. Start using the MCP server through the client you configured (Claude Code or Codex).

Tips and common notes:

• Ensure you provide a valid WILDCARD_API_KEY for your DeepContext integration. Without it, requests will fail to authenticate.
• The MCP server is designed to work with TypeScript and Python codebases, leveraging Tree-sitter for robust parsing and semantic extraction.
• Indexing is incremental: the server tracks file hashes and modification times to reprocess only changed files, which helps large repos stay responsive.
• If you plan self-hosting, you may need additional keys for vector storage or reranking services (e.g., Turbopuffer and Jina) depending on your deployment setup. See the self-hosting notes in the repository for details.
• When indexing, consider excluding test files, generated code, and build outputs to keep the index focused on source code.
• You can adjust the MCP integration flow weights between vector similarity and BM25 in your deployment to tune results for your codebase.