Contextstream MCP server
{
"mcpServers": {
"contextstream": {
"type": "http",
"url": "https://mcp.contextstream.io/mcp?default_context_mode=fast&src=registry"
}
}
}{
"mcpServers": {
"contextstream": {
"type": "http",
"url": "https://mcp.contextstream.io/mcp?default_context_mode=fast&src=registry"
}
}
}{
"mcpServers": {
"contextstream": {
"type": "http",
"url": "https://mcp.contextstream.io/mcp?default_context_mode=fast&src=registry"
}
}
}{
"servers": {
"contextstream": {
"type": "http",
"url": "https://mcp.contextstream.io/mcp?default_context_mode=fast&src=registry"
}
}
}{
"mcpServers": {
"contextstream": {
"type": "http",
"url": "https://mcp.contextstream.io/mcp?default_context_mode=fast&src=registry"
}
}
}file .mcp.json (in your project root)file claude_desktop_config.jsonfile ~/.cursor/mcp.jsonfile .vscode/mcp.jsonfile ~/.codeium/windsurf/mcp_config.jsonversion 0.5.35
What it does
ContextStream is a remote-hosted MCP server that provides persistent memory across coding sessions, semantic code search, and project context management for AI coding assistants. It works with Claude Code, Cursor, VS Code + Copilot, and other MCP clients to help coding agents retain decisions and architectural knowledge between sessions.
- Retrieve relevant code context and architectural decisions from previous sessions without re-indexing the repository each time.
- Search code by semantic meaning rather than keyword matching to find related functionality across a project.
- Maintain a persistent log of project decisions and lessons learned so coding agents avoid repeating mistakes.
Tools this server exposes
| Tool | Does |
|---|---|
| semantic_code_search | Search code by meaning across the linked project repository with high recall. |
| project_context_retrieval | Retrieve relevant project architecture, dependencies, and decision history for the current coding task. |
| persistent_session_memory | Store and restore multi-session coding context and lessons learned across chat sessions. |
Before you enable it
An MCP server runs with whatever access you give it. Start with a development account rather than production, and check which tools the server exposes before you let an agent call them unattended.