The MCP Memory Server for AI Coding Agents
ContextForge is an MCP server that gives any AI tool persistent memory. Built on the Model Context Protocol standard by Anthropic — works with Claude Code, Cursor, GitHub Copilot, ChatGPT, Windsurf, and Claude Desktop.
MCP is powerful but stateless
MCP gives your AI tools access to external data, but most MCP servers are read-only. There's no standard way to give your AI a memory that grows over time.
A purpose-built memory MCP server
ContextForge is designed from the ground up for AI memory. Save knowledge with semantic embeddings, search by meaning, organize by projects, share with your team — all through MCP.
Setup in under 3 minutes
Sign up at contextforge.dev
Free tier includes 200 documents, 500 queries/month, and 1 project. No credit card required.
Install the MCP server
One command works for any MCP-compatible tool:
npx contextforge-mcpConnect to your AI tool
Configure your MCP client (Claude Code, Cursor, Copilot, etc.) to point to the ContextForge MCP server. See our docs for tool-specific setup.
What you can do
Universal compatibility
Works with every MCP client. Build your knowledge base once, access it from Claude Code, Cursor, Copilot, ChatGPT — any tool that speaks MCP.
Semantic search
Vector embeddings power the search. "How do we handle payments?" finds your "Invoice Workflow" doc because the meanings match.
Knowledge graph
Link related items with typed relationships. When you query one topic, related knowledge surfaces automatically.
Open source MCP client
The npm package contextforge-mcp is open source and lightweight. Install in seconds, no server-side dependencies.
Secure by default
Row-level security at the database level. API keys hashed with SHA-256. Your data is yours — never used for AI training.
Common questions
What is MCP (Model Context Protocol)?
MCP is an open standard by Anthropic that lets AI tools connect to external services. Think USB for AI — one standard protocol, any tool. ContextForge uses MCP to give your AI persistent memory.
How is this different from the official MCP memory server?
The official @modelcontextprotocol/server-memory stores entities locally as JSONL. ContextForge is a hosted MCP memory server with semantic search, multi-device sync, team collaboration, Git integration, and a web dashboard. You can also import from the official MCP memory format.
Can I self-host ContextForge?
The MCP client (contextforge-mcp on npm) is open source. The backend is hosted at contextforge.dev. Self-hosting the backend isn't supported, but you can export your data anytime in JSON or Markdown.
Which MCP tools does ContextForge expose?
Memory operations (ingest, query, list, delete), project/space management, knowledge graph relationships, Git integration (sync commits/PRs), task tracking, snapshots, and import/export. See the docs for the full tool reference.
Ready to give MCP memory?
Free tier available. Setup takes under 3 minutes.