Skip to content

llms-install.md

Latest commit

 

History

History
60 lines (43 loc) · 2.35 KB

llms-install.md

File metadata and controls

60 lines (43 loc) · 2.35 KB
Raw

ContextStream MCP Server — LLM Installation Guide

This guide is for AI agents (Cline, Claude, Cursor agents, etc.) installing the ContextStream MCP server on a user's behalf.

ContextStream gives AI coding assistants persistent memory and cross-session learning: semantic code search, knowledge graphs, decisions, lessons, and session transcripts that survive across sessions and tools.

Prerequisites

  • A ContextStream API key. The user can create a free account at https://contextstream.io and copy the key from the dashboard (Settings → API Keys). If the user does not have a key yet, pause and ask them to create one — do not invent a value.
  • For Option 1 (npx): Node.js 18+ available on PATH.

Option 1 — npx (stdio, recommended for Cline)

Add this to the MCP settings file (for Cline: cline_mcp_settings.json):

{
  "mcpServers": {
    "contextstream": {
      "command": "npx",
      "args": ["--prefer-online", "-y", "@contextstream/mcp-server@latest"],
      "env": {
        "CONTEXTSTREAM_API_URL": "https://api.contextstream.io",
        "CONTEXTSTREAM_API_KEY": "USER_API_KEY_HERE"
      }
    }
  }
}

Replace USER_API_KEY_HERE with the user's real API key.

Option 2 — Hosted remote (streamable HTTP, zero install)

For clients that support remote MCP servers with OAuth (VS Code, Claude, etc.), no local install or API key in config is needed:

{
  "servers": {
    "contextstream": {
      "type": "http",
      "url": "https://mcp.contextstream.io/mcp?default_context_mode=fast"
    }
  }
}

The endpoint uses built-in OAuth — the client will open a browser window for the user to authorize on first use.

Verify the installation

After the server connects, call the init tool (no arguments needed beyond the project folder path if available). A successful response starts with Session ready for workspace: …. If tools do not appear, restart the MCP client.

Troubleshooting

  • 401 Unauthorized: CONTEXTSTREAM_API_KEY is missing, mistyped, or revoked. Ask the user to re-copy it from the dashboard.
  • npx download issues: run npx -y @contextstream/mcp-server@latest --version in a terminal to pre-warm the cache.
  • Corporate proxy/firewall: the server needs outbound HTTPS to api.contextstream.io (Option 1) or mcp.contextstream.io (Option 2).

More documentation: https://contextstream.io/docs/mcp