World Model MCP brings long-term memory to Claude Code
The developer released World Model MCP v0.9.1, which builds a temporal knowledge graph for AI agents. The system lets Claude Code, Cursor, and others recall fixes, prevent hallucinations, and avoid code repetition.
The update includes a +10.2 SWE-bench Verified accuracy gain. The system analyzes user edits, saves them as constraints, and applies them in future sessions. It resolves contradictions based on confidence weights and fact lifetimes.
The tool runs as a SQLite-based local memory layer. Installation is via pip and takes minutes to set up. It supports GitHub, Notion, and Slack via the MCP standard.
Related stories
- In Claude Code, one CLAUDE.md silently kills your AGENTS.md
- Claude Code matches Codex on SWE-Bench Pro at twice the cost
- Claude Code 2.1.259 hands MCP servers to the whole org
- Claude Code stores MCP OAuth tokens in plaintext on Linux
- Context7 cuts Claude Code input tokens by roughly 99%
- A harness pushes Claude Opus 5 past 30% on ARC-AGI-3
Comments
No comments yet. Be the first.
Join the conversation
Sign in with Google to leave a comment. Your name and avatar come from your Google profile, and the comment appears after moderation.