AI coding agents forget your project between sessions. ContextForge fixes that with persistent memory via MCP — it auto-recalls your project when a session starts and saves decisions as you work, no manual commands. Semantic search understands your code and your git history. Works with Claude, Cursor, Copilot, Windsurf & any MCP client. Free tier, 3-min setup.

ContextForge gives Claude Code a real long-term memory. I build CrawlForge (a web-scraping MCP server) mostly on my own, and I used to start every session re-explaining the project: which branch ships, which Stripe quirks to avoid, what I'd already ruled out. Now Claude checks ContextForge first and picks up where we left off. It feels like working with a teammate who was there yesterday.

It really proved itself while I prepared my own Product Hunt launch over many sessions: copy, timing, checklists, videos. One question brought all of it back each time. It also pairs well with CrawlForge. CrawlForge does the web research, ContextForge keeps the conclusions, and weeks later Claude still knows what I found and decided. Setup was just an MCP config and an API key.

Tighter project scoping. When I work across several repos, a task sometimes gets attached to the wrong project. I'd also love a one-click import of existing Claude Code memory files, so new users can bring their history with them on day one.

Claude Code's built-in memory: CLAUDE.md plus local memory files. I still use it for project rules, but it's tied to one machine and one folder, and you have to load files by hand to recall anything. ContextForge adds semantic search, spaces to organize by area, and tools to correct or retire stale memories. I've ended up using ContextForge as the source of truth.

ContextForge

Developer Tools Artificial Intelligence Productivity MCP

Developer Tools Artificial Intelligence Productivity MCP

ContextForge Launches

ContextForge Launches

What's great

What needs improvement

vs Alternatives

What's great

What needs improvement

vs Alternatives