ContextStream

ContextStream gives AI coding agents persistent project memory and semantic code search across sessions.

Connect your assistant to ContextStream’s hosted production MCP to retrieve the decisions, lessons, repo guardrails, task history, and code understanding your team has already built up. Future sessions start with the right project context instead of relearning the repository from scratch.

Hosted production MCP

Use the hosted endpoint:

https://mcp.contextstream.io/mcp?default_context_mode=fast

The hosted service is the current production MCP. It is operated by ContextStream and uses the current Rust runtime. You do not need to install or run the legacy npm MCP server.

What agents can do

• Retrieve relevant project context before editing code
• Search code semantically and by exact text, symbols, paths, and patterns
• Recall decisions, lessons, plans, sessions, and repo guardrails
• Preserve corrections and implementation knowledge for future sessions
• Share project context across supported MCP clients and teammates

Setup

The fastest setup path is:

curl -fsSL https://contextstream.io/scripts/mcp.sh | bash

The setup detects supported assistants, signs you in, configures the production MCP connection, and verifies the result.

For manual editor configuration and OAuth instructions, see:

https://contextstream.io/docs/editors

Supported clients

ContextStream works with MCP-compatible coding assistants and editors, including Claude Code, Cursor, Codex CLI, Cline, Windsurf, VS Code with GitHub Copilot, and other MCP clients.

Privacy and access

ContextStream scopes access by workspace and project. The hosted MCP requires authentication; project data is not exposed through an unauthenticated public endpoint.