mcp

ContextStream MCP Server is a powerful AI memory and context engine designed to enhance your coding assistant with semantic understanding, smart context delivery, and a connected knowledge graph. It exposes a suite of tools that your AI can invoke to load workspace context, search code, capture decisions, and build a persistent memory of your project. The available tools include: init to load your workspace context instantly, context to deliver relevant context for each message, search for semantic or keyword-based discovery, session to capture decisions and lessons automatically, memory to build a knowledge graph of the project, graph to map dependencies and analyze impact, project to index the codebase for semantic understanding, media to index and search multimedia assets, and integration to query GitHub, Slack, and Notion directly. You don’t need to manually wire these together—the MCP server handles orchestration, while you focus on building intelligent prompts and workflows that leverage these capabilities. The setup wizard (invoked via npx @contextstream/mcp-server@latest setup) configures authentication, runtime settings, editor integrations, and optional hooks to supercharge your workflow.

To use the MCP server in your development environment, start with the recommended setup command provided in the README, typically:

npx @contextstream/mcp-server@latest setup

After setup, you can configure tools in your editor integrations (Claude Code, Cursor/Claude Desktop, VS Code, Copilot CLI, etc.) by pointing them at the MCP server using the provided mcpServers configuration. You’ll usually provide an API key via CONTEXTSTREAM_API_KEY for authenticated access. The server then responds to your prompts with context-aware results, leveraging semantic search, lessons, and the knowledge graph to keep your AI informed and efficient.

Prerequisites:

• Node.js (recommended: v14+ or newer; check project requirements)
• npm (comes with Node.js) or pnpm/yarn if preferred
• Internet access to install packages

Quick start (recommended):

1. Ensure Node.js and npm are installed and up to date.
2. Run the setup wizard directly in your project (no global install needed):

npx @contextstream/mcp-server@latest setup

This wizard will guide you through authentication, configuration, and editor integration.

Optional alternative — local/global install (if you prefer):

# Install globally (optional)
npm install -g @contextstream/mcp-server

# Run the setup via the installed CLI (name may vary; follow package docs if needed)
mcp-server setup

Configuration notes:

• If you plan to run without interactive setup, you can provide a manual configuration JSON (see example in this README) and place it where your editor integrations expect it.
• Remember to set your API key in the environment variable CONTEXTSTREAM_API_KEY when required by your setup.

Verification:

• After setup, verify that the MCP server is reachable from your editor integration by testing a simple command (e.g., init) through your tool’s command palette or integration UI.

Tips and common notes:

• Environment variables: CONTEXTSTREAM_API_KEY is commonly required for authenticated access; store securely and avoid committing keys to version control.
• Editor integrations: Most integrations (Claude Code, Cursor/Desktop, VS Code, Copilot CLI) expect a configuration snippet pointing to your MCP server with command 'npx' and args including the package name. You can customize the mcpServers section as needed.
• Setup wizard: The setup step configures authentication, workspace indexing, and optional hooks. Re-run if you change your workspace or need to reconnect integrations.
• If you encounter network or permission issues, ensure your npm/Node.js environment permits executing npx commands and that your network allows access to npm registries.
• When using the manual JSON configuration, ensure the JSON structure aligns with your editor’s expected config (e.g., Claude, Copilot CLI, VS Code).
• For production use, consider securing API keys and using a dedicated environment for your MCP server to avoid conflicts with local developer tooling.