# contextstream/mcp-server [Health: Active]

**Category:** 🧠 Knowledge & Memory  
**Repository:** https://github.com/contextstream/mcp-server  
**GitHub Stars:** 44  
**npm Downloads (last month):** 7491  
**Views:** 2  
**Installs:** 0  
**Upvotes:** 0  
**Directory Page:** https://allmcps.com/mcp/contextstream-mcp-server

## Description
Universal persistent memory for AI coding tools. Semantic code search, knowledge graphs, impact analysis, and decision tracking. 90.0% on LongMemEval-S. Works across Cursor, Claude Code, Windsurf, and any MCP client. npx -y @contextstream/mcp-server

## Tools
Capabilities this server exposes over MCP:

- **search** — semantic, hybrid, keyword, pattern, exhaustive, refactor modes
- **memory** — events, decisions, docs, runbooks, tasks, todos, diagrams, transcripts
- **session** — capture decisions & lessons, recall past sessions, plans, retroactive capture
- **qa** — grounded Q&A over your workspace knowledge base, with citations
- **graph** — dependencies, impact analysis, circular deps, unused code
- **capsule** — portable, shareable context snapshots for agent handoffs
- **entity** — tickets, incidents, releases, sprints, OKRs, risks
- **skill** — reusable instruction + action bundles, portable across tools
- **media** — index and search images, video, audio, and documents

## Claude Desktop Quick Installation
Install path detected from listing signals. Uses `npx` (confidence: high):

```json
"mcpServers": {
  "mcp-server": {
    "command": "npx",
    "args": ["-y","@contextstream/mcp-server@latest"],
    "env": {
      "CONTEXTSTREAM_API_KEY": ""
    }
  }
}
```

**Requires environment variables:** `CONTEXTSTREAM_API_KEY` — the values above are empty placeholders; fill in real credentials before running (see the repository for what each one is for).

## Documentation

## What contextstream/mcp-server does

The contextstream/mcp-server MCP server gives coding agents access to persistent project context across sessions. It can connect code, documentation, and conversations so an agent can retrieve relevant files, saved decisions, lessons, plans, and other workspace knowledge. This places it in the 🧠 Knowledge & Memory category, with additional tools for code analysis and work tracking.

Search supports semantic, hybrid, keyword, pattern, exhaustive, and refactor modes. The memory functions cover events, decisions, documents, runbooks, tasks, todos, diagrams, and transcripts. Session features capture decisions and lessons, recall earlier sessions, manage plans, and support retroactive capture. Grounded Q&A can answer over the workspace knowledge base and include citations.

## How it works

The server exposes MCP tools through a local installation or a hosted endpoint. Local setup indexes eligible project source contents for hosted search; transcript saving and local Git metadata capture are enabled by default but can be disabled. A hosted connection does not synchronize a local checkout by itself.

Beyond retrieval, the contextstream/mcp-server MCP server can inspect dependency relationships, analyze impact, identify circular dependencies, and find unused code. Capsules provide portable context snapshots for agent handoffs. Entity tools represent tickets, incidents, releases, sprints, OKRs, and risks, while skills package reusable instructions and actions. Media indexing and search covers images, video, audio, and documents.

## Setup and configuration

On macOS or Linux, the repository documents an installation script available through `curl`; Windows uses a PowerShell equivalent. The npm option is `npx -y @contextstream/mcp-server@latest setup` and requires Node.js 20 or newer. The launcher downloads a platform-specific Rust binary and verifies its SHA-256 checksum. Pin an exact package version for production automation.

Interactive onboarding connects an account, project, and supported editor. For scripted setup, `setup --yes` can read `CONTEXTSTREAM_API_KEY` or saved credentials; accounts with multiple workspaces also need a workspace ID. `setup --dry-run` previews configuration, while `doctor --scope=all --only-configured` diagnoses configured connections. Account-level learning can be reviewed or changed with the configure command.

A hosted Streamable HTTP endpoint is available at `https://mcp.contextstream.io/mcp?default_context_mode=fast` for clients that support Streamable HTTP and OAuth. The repository contains the MIT-licensed Rust server, client, editor setup, and release tooling; the hosted backend is separate.

## Tools and capabilities

- Search code and workspace content using semantic, hybrid, keyword, pattern, exhaustive, or refactor modes.
- Store and retrieve project knowledge, including decisions, runbooks, tasks, transcripts, and diagrams.
- Capture or recall session decisions, lessons, and plans.
- Ask cited questions against the workspace knowledge base.
- Analyze dependencies, change impact, circular dependencies, and unused code.
- Package handoff context into capsules and manage entities, skills, and media.

## Limitations and notes

The contextstream/mcp-server MCP server does not recover every undocumented decision merely by indexing code; those decisions need to be captured. Indexing sends eligible source contents to ContextStream for hosted search, so review the project’s data-handling controls before connecting sensitive repositories. The published benchmark reports a 95 ms median code-search result for its disclosed configuration, but that figure is not a latency guarantee.

_Full upstream README: https://allmcps.com/mcp/contextstream-mcp-server/readme_