SaravananJaichandar/world-model-mcp

Overview

An MCP server that builds a persistent world model of a codebase to reduce hallucinations, repeat mistakes, and regressions when working with Claude Code. It runs in Python and provides structured context to improve the consistency of AI-generated code suggestions.

Best for

Best for
Developers using Claude Code who need more reliable, context-aware code generation

Use cases

• Integrating with Claude Code to maintain accurate context across sessions
• Preventing repeated coding errors by tracking past mistakes in the codebase
• Avoiding regressions by ensuring new code aligns with existing patterns and constraints

How to use

Install

pip install -U world-model-mcp && world-model demo

Tools exposed

• query_fact
• record_event
• validate_change
• get_constraints
• record_correction
• get_related_bugs
• seed_project
• ingest_pr_reviews
• WORLD_MODEL_AUDIT_LOG
• WORLD_MODEL_DB_PATH
• WORLD_MODEL_TELEMETRY
• ANTHROPIC_API_KEY

Tested with

Claude Desktop, Claude Code, Cursor, Windsurf, Cline, Continue, VS Code

Example client config

WORLD_MODEL_AUDIT_LOG=on\nWORLD_MODEL_VERIFICATION_BACKEND=openai-compatible

Notes

An MCP server that builds a persistent world model of a codebase to reduce hallucinations, repeat mistakes, and regressions when working with Claude Code. It runs in Python and provides structured context to improve the consistency of AI-generated code suggestions.

4 stars on GitHub. Last updated 2026-06-05. Licensed MIT.

Use cases

• Integrating with Claude Code to maintain accurate context across sessions
• Preventing repeated coding errors by tracking past mistakes in the codebase
• Avoiding regressions by ensuring new code aligns with existing patterns and constraints

Pros

• Reduces LLM hallucinations by grounding responses in a live codebase model
• Production-ready design suitable for ongoing development workflows
• Explicitly targets common pain points like error repetition and regressions

Cons

• Tightly coupled to Claude Code, limiting broader LLM tool compatibility
• Requires a running MCP server and Claude Code setup, adding infrastructure overhead
• Stars count suggests a young or niche repository with limited community validation

Indexed from awesome-mcp-servers-punkpeye and enriched against its public facts.