Cursor-Cortex: Structured Memory for AI Assistants

Works with Cursor, Claude Code, and any MCP-compatible AI assistant.

Solve AI assistant memory loss and context confusion. Cursor-Cortex gives your AI the structured thinking and long-term memory it needs to actually help you build software.

🧠 The Problem with AI Assistants

LLM-based AI assistants are amazing at generating code and explaining concepts, but they struggle with:

• ❌ Memory Loss: "What were we working on yesterday?"

• ❌ Context Switching: "Wait, what is this project about again?"

• ❌ Agent Failures: Getting lost, hallucinating, losing track of goals

• ❌ No Structure: Jumping between ideas without systematic thinking

• ❌ Knowledge Gaps: No access to company-specific or project-specific insights

Result: You spend more time explaining context than actually solving problems.

Related MCP server: ContextStream MCP Server

🎯 The Cursor-Cortex Solution

Cursor-Cortex provides your AI with structured external memory:

📋 Context Files → "What is this project about?"

Your AI always knows the project goals, architecture decisions, and current focus.

📝 Branch Notes → "What have we been working on?"

Continuous development log so your AI can pick up exactly where you left off.

🧠 Tacit Knowledge → "How do we solve this type of problem?"

Company-specific solutions, patterns, and hard-won insights that LLMs don't know.

✅ Checklists → "What's our systematic approach?"

Structured thinking frameworks that keep AI (and you) organized and thorough.

💡 Before vs After

❌ Without Cursor-Cortex:

You: "Help me continue the authentication work"
AI: "I need more context. What authentication system? What issues were you facing?"
You: "Ugh, let me explain everything again..."

✅ With Cursor-Cortex:

You: "Help me continue the authentication work"  
AI: *reads context + branch notes + knowledge base*
"I see you're implementing JWT refresh tokens and hit CORS issues yesterday. 
Based on our company knowledge doc about API security, let's check the middleware..."

🚀 Core Benefit

Turn your AI from a smart intern into an experienced teammate who remembers everything, follows structured thinking, and has access to institutional knowledge.

⚡ Complete Setup Guide

1. Install Dependencies

npm install --legacy-peer-deps

Note: The --legacy-peer-deps flag is required due to TensorFlow.js peer dependency conflicts between v3.x and v4.x versions.

2. Get Full Path

pwd
# Copy the output - you'll need it for configuration

3. Vector Search Setup (Optional)

For AI-powered semantic search capabilities:

System Requirements:

• Node.js 18+

• 2GB RAM for vector models

• 500MB disk space

Setup Steps:

# Step 1: Generate embeddings for all knowledge files
node generate-all-embeddings-cpu.js

# This creates vector embeddings for:
# - All tacit knowledge documents
# - All branch notes
# - All context files
# Duration varies by document count and system specs

# Step 2 (Optional): Force regenerate if content changes
node generate-all-embeddings-cpu.js --force

# Step 3 (Optional): View progress during generation
node generate-all-embeddings-cpu.js --verbose

Why This Matters:

• Without embeddings, semantic search falls back to text search

• Pre-computed embeddings make searches 100x faster

• Generate once, use for all semantic queries

• Embeddings stored in ~/.cursor-cortex/embeddings/

4. Configure Cursor MCP

Create or edit ~/.cursor/mcp.json:

"cursor-cortex": {
      "command": "node",
      "args": ["/YOUR/FULL/PATH/TO/cursor-cortex/index.js"],
      "tools": [
        "update_branch_note", 
        "add_commit_separator",
        "read_branch_notes",
        "update_context_file", 
        "read_project_context",
        "read_branch_context", 
        "list_context_files", 
        "generate_commit_message", 
        "generate_jira_comment", 
        "create_tacit_knowledge",
        "read_tacit_knowledge", 
        "search_branch_notes",
        "generate_embeddings",
        "create_completion_checklist",
        "read_checklist",
        "update_checklist",
        "sign_off_checklist",
        "archive_branch_note",
        "clear_branch_note",
        "list_all_branch_notes",
        "enhanced_branch_survey",
        "comprehensive_knowledge_search",
        "construct_project_narrative",
        "timeline_reconstruction",
        "context_sync_guidance",
        "analyze_documentation_gaps",
        "migrate_context_files",
        "request_critical_thinking_space",
        "check_critical_thinking_status",
        "request_thinking_guidance",
        "add_perspective",
        "request_synthesis_space",
        "request_synthesis_step_guidance",
        "complete_synthesis_step",
        "generate_context_zip",
        "unpack_context"
      ],
      "stdio": true,
      "env": {
        "DEBUG": "true"
      }
    },

Replace /YOUR/FULL/PATH/TO/cursor-cortex/ with the output from step 2

Note: This configuration is for Cursor-Cortex v1.2.0+. If you're using an older version, some tools may not be available. Check your package.json version and consider updating.

5. Restart Cursor

Close and reopen Cursor completely.

Claude Code Installation

Create or edit .claude/settings.json in your project root (or ~/.claude/settings.json globally):

{
  "mcpServers": {
    "cursor-cortex": {
      "command": "node",
      "args": ["/YOUR/FULL/PATH/TO/cursor-cortex/index.js"]
    }
  }
}

Replace /YOUR/FULL/PATH/TO/cursor-cortex/ with the output from step 2, then restart Claude Code.

6. Enable MCP Server

1. Go to Cursor Settings (Cmd/Ctrl + ,)

2. Navigate to Features → Model Context Protocol

3. Find cursor-cortex and toggle it ON

4. You should see: Storage directory created at /Users/yourname/.cursor-cortex

7. Test It

In any Cursor chat, try:

"Update my branch notes: Testing Cursor-Cortex setup"

If it works, you'll see: Successfully updated branch note with: "Testing Cursor-Cortex setup"

8. Knowledge Explorer (Visual Graph)

Once everything is set up, explore your knowledge base as an interactive network graph:

npm run explorer

Then open http://localhost:4000 in your browser.

Features:

• Three-tier visualization: Projects, branch notes, and tacit knowledge as interconnected nodes

• Similarity-driven layout: Semantically similar projects cluster together using embedding cosine similarity

• Neural pulse animation: Watch signals propagate through your knowledge network in real-time

• Activity-based colors: Brighter nodes = more recently updated projects

• Click to explore: Click any node to see details, similar projects, and tags

• Depth toggle: Switch between Compact (projects only), Branches, and Full views

🔧 Troubleshooting

Common Issues:

❌ "Tool not found" error

• Restart Cursor completely after configuration

• Verify MCP server is enabled in settings

• Check file path in mcp.json is correct (no typos)

❌ MCP configuration not working

• Validate JSON syntax at jsonlint.com

• Ensure commas and quotes are correct

• Check file permissions on mcp.json

❌ Vector search not working

• Run npm install to ensure TensorFlow.js is installed

• Check system has 2GB+ RAM available

• Try text search mode if vector search fails

Need Help? Create an issue with your error message for quick support.

9. Start Using

• "Show my uncommitted work" - See what you've been working on

• "Generate a commit message" - Create commits from your notes

• "Create tacit knowledge document" - Save solutions for later

📖 Basic Usage

Capture Your Thoughts

As you code, tell your AI assistant:

"Update branch notes: Fixed the login validation bug by adding proper error handling 
for expired tokens. Had to modify the JWT middleware."

Stay Oriented

When you return to work:

"Show me my uncommitted work"

Smart Commits

When ready to commit:

"Generate a commit message"

AI reads all your notes and creates: "Fix JWT token validation and error handling"

🏗️ Project Structure

Cursor-Cortex organizes your knowledge into:

• 📝 Branch Notes: What you're working on right now

• 📋 Context Files: Project documentation and decisions

• 🧠 Knowledge Base: Solutions and insights for future reference

• ✅ Checklists: Track project completion

All stored in ~/.cursor-cortex/ and accessible through AI.

🎓 Power User Features

Once you're comfortable with basics, explore these advanced capabilities:

Cross-Project Knowledge Search

"Search all my projects for JWT authentication solutions"

Automatic Commit Tracking

node setup-hooks.js  # Auto-separates pre/post commit work

Project Checklists

Track requirements, testing, and sign-offs:

"Create a completion checklist for the user auth feature"

Team Knowledge Sharing

Create reusable solution documents:

"Create tacit knowledge document about our JWT implementation approach"

Knowledge Archaeology & Project Intelligence

Analyze and synthesize project knowledge:

"Survey all my branch documentation and show completeness scores"
"Construct a project narrative for the user authentication system"
"Show me a timeline of all development across projects"
"Analyze this folder structure for documentation gaps"

🛠️ All Available Tools

• update_branch_note - Add entries to your development log

• read_branch_notes - Default is the full file (including COMMIT separators); mode=uncommitted or uncommittedOnly=true for work since the last commit; optional beforeDate, afterDate, or commitHash to filter

• add_commit_separator - Mark commit boundaries (auto via git hooks)

• generate_commit_message - AI creates commits from your notes

• update_context_file - Document project goals and decisions

• read_project_context - Read branch-agnostic project context only

• read_branch_context - Read branch-specific context only

• list_context_files - Browse all project docs with cross-project warnings

• create_tacit_knowledge - Document solutions for future reference

• read_tacit_knowledge - Search knowledge across all projects with tags and content search

• search_branch_notes - Search branch notes across projects with semantic search

• generate_embeddings - Generate or regenerate vector embeddings for semantic search

• comprehensive_knowledge_search - Global semantic search across ALL Cursor-Cortex knowledge

• create_completion_checklist - Track requirements and deliverables

• read_checklist - View progress

• update_checklist - Mark items completed (manual or auto-detect)

• sign_off_checklist - Formal approval with signatures

• archive_branch_note - Archive completed work

• clear_branch_note - Reset branch documentation

• migrate_context_files - Migrate existing context files to Smart Hybrid Context System

• list_all_branch_notes - View all branch notes across all projects with priority ordering

• enhanced_branch_survey - Comprehensive analysis of documentation with completeness scoring and relationship mapping

• construct_project_narrative - Weave scattered technical details into coherent production stories

• timeline_reconstruction - Extract chronological timeline data from branch notes and commit separators

• context_sync_guidance - Get contextual guidance for syncing project contexts based on timeline data

• analyze_documentation_gaps - Analyze folder structure to identify documentation needs and auto-create checklists

• request_critical_thinking_space - Create systematic analysis workspace using Six Thinking Hats methodology

• check_critical_thinking_status - Check completion status of critical thinking analysis

• request_thinking_guidance - Get guidance for specific Six Thinking Hats perspectives

• add_perspective - Add specific perspective analysis to critical thinking workspace

• request_synthesis_space - Load perspectives into synthesis workspace for integration

• request_synthesis_step_guidance - Get detailed guidance for synthesis steps

• complete_synthesis_step - Complete specific synthesis step and update process

• generate_context_zip - Package knowledge into portable ZIP archive for sharing between users/environments

• unpack_context - Import shared context package with safe conflict resolution

🎛️ Advanced Configuration

Git Integration

Automatic commit separators require the MCP server to be running:

node setup-hooks.js  # Install git hooks

CLI Interface

Direct access without AI:

node cursor-cortex-cli.js  # Interactive menu

🤝 Contributing

Found a bug or have an idea? Open an issue or submit a pull request — contributions are welcome!

Author: Manuel Flores-Ramirez

📄 License

This project is licensed under the Parachute Public License (PPL) v1.0.

The PPL is an ethical license that includes restrictions against use by government, military, surveillance, and other harmful entities. For full details and rationale, see parachute.pub.

© 2025 Manuel Flores-Ramirez.