ContextStream

ContextStream MCP Server provides a persistent memory layer for your AI coding tools, enabling remembered decisions, preferences, and context across tools and sessions. It integrates with multiple editors and MCP clients to keep your workspace context available wherever you work, boosting onboarding, consistency, and productivity.

How to use

You set up a local MCP server integration in your editor or tool, then connect your MCP client to it. Once configured, your AI remembers decisions, preferences, and context across sessions and tools, so you can refer back to past conversations, decisions, and lessons without re-explaining.

How to install

Prerequisites: Node.js and npm must be installed on your system. If you don’t have Node.js, install it from the official Node.js website or use your operating system’s package manager.

# Install the MCP server globally
npm install -g @contextstream/mcp-server

# Or run via npx (recommended for MCP configurations)
npx @contextstream/mcp-server

Configuration and usage notes

Configure your MCP client to run the MCP server locally with the following stdio configuration. This uses npx to start the MCP server and passes the required environment variables for API access.

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

Codex CLI and other editors configuration

If you use Codex CLI, place a similar configuration where your Codex settings reside. The code example shows how to point Codex to the MCP server and provide the API credentials.

[mcp_servers.contextstream]
command = "npx"
args = ["-y", "@contextstream/mcp-server"]

[mcp_servers.contextstream.env]
CONTEXTSTREAM_API_URL = "https://api.contextstream.io"
CONTEXTSTREAM_API_KEY = "your_api_key"

First session and memory behavior

Open a new conversation in your editor and tell your AI to initialize memory for your project. You can then set preferences, such as coding style or framework choices, and start capturing decisions. The memory persists across sessions and tools.

Example prompts you can start with once the integration is active:

You: Initialize session and remember I prefer functional React components

You: Capture decision: Use PostgreSQL for the user database

Security considerations

All data is encrypted at rest and access is controlled via workspace permissions and API keys. You can delete your data at any time.

Available tools

session_init

Initialize a new memory session and automatically load recent context.

context_smart

Fetch relevant context for any incoming message.

session_remember

Store a memory item via natural language command.

session_recall

Retrieve remembered decisions or preferences.

session_capture

Persist decisions, insights, and preferences from the current session.

session_capture_lesson

Capture mistakes as lessons to prevent repeating them.

session_get_lessons

Retrieve previously captured lessons.

search_semantic

Find code or docs by meaning rather than exact keywords.

search_hybrid

Combine semantic search with keyword matching for code.

graph_dependencies

Visualize what depends on what in your codebase.

graph_impact

Assess change impact before making edits.

graph_call_path

Trace execution paths across modules.

graph_unused_code

Identify dead or unused code.

ai_context

Build LLM-ready context for prompts.

ai_context_budget

Manage token budget for context.

ai_plan

Generate development plans from high-level goals.

ai_tasks

Break work into actionable tasks.