Setup Guide

npm install -g @anthropic-ai/claude-code

{
  "mcpServers": {
    "eon-memory": {
      "type": "streamable-http",
      "url": "https://mcp.ai-developer.ch/mcp/",
      "headers": {
        "Authorization": "Bearer eon_YOUR_API_KEY"
      }
    }
  }
}

# EON Memory -- Your AI Knowledge System

You have access to EON Memory, a persistent knowledge system with truth validation.
Every memory you create is automatically validated through 15 truth tests and scored for ethical alignment.

## Your MCP Tools

You have 16 tools available through the EON MCP server:

### Core Memory
- eon_create -- Save validated knowledge (title, content, project_id, category)
- eon_search -- Semantic search (finds by meaning, not just keywords)
- eon_list -- Browse memories with pagination
- eon_get -- Retrieve a specific memory by ID
- eon_update -- Update and re-validate a memory

### Extended
- eon_delete -- Soft delete (recoverable)
- eon_projects -- List all projects with memory counts
- eon_similar -- Find memories similar to a given one
- eon_session_start -- Begin a work session
- eon_session_end -- End session, save summary

### Goals
- eon_goals_create -- Create goals with priority and deadline
- eon_goals_list -- List goals by status/priority
- eon_goals_update -- Update goal progress

### System
- eon_health -- System health check
- eon_stats -- Usage statistics and quality metrics
- eon_scan -- Auto-create memories from code directories

## How to Work

1. Start of session: Use eon_search to find relevant existing knowledge
2. During work: Save important decisions, context, and learnings with eon_create
3. End of session: Use eon_session_end to save a session summary

## Categories for Memories
- update -- Progress updates, status changes
- context -- Background information, architecture decisions
- decision -- Important decisions with reasoning
- error -- Bugs found, errors encountered, lessons learned
- roadmap -- Plans, milestones, future work

## Quality Guidelines
- Write clear, specific titles (min 3 chars)
- Include enough detail in content (min 20 chars)
- Organize by project_id for clean search results
- Search before creating to avoid duplicates
- If a memory gets a low quality score, improve the content

## Memory Detail Guidelines (CRITICAL)

When creating memories, think: "Could another AI use this to WRITE CODE in 6 months?"

For API endpoints, ALWAYS include:
- HTTP method, path, full parameter list with types
- Response format with example JSON and all status codes (200, 400, 429, 500)
- Import chain (which files this endpoint depends on)
- The step-by-step execution flow
- Code template for creating similar endpoints

For database schemas, ALWAYS include:
- Full CREATE TABLE with column types, constraints, defaults
- Foreign keys, indexes, unique constraints
- Example queries for common operations

For components/functions, ALWAYS include:
- Full function signature with parameter types and return type
- Dependencies, imports, and side effects
- State management approach

For configuration/deployment, ALWAYS include:
- All environment variables with defaults
- Port mappings, service names, startup commands

BAD example (4/10 usefulness): "15 API routes exist with rate limiting"
GOOD example (9/10 usefulness): "POST /api/vote - no body needed, session from cookie, returns {success: true, count: 42}, 429 on rate limit, imports getDb, applySecurity, getSessionId"

## X-Ethics Principles
Every memory is scored against four ethical pillars:
- Truth (W): Is the content truthful and non-deceptive?
- Freedom (F): Does it respect autonomous choice?
- Justice (G): Does it treat all subjects with fairness and dignity?
- Service (D): Does it create genuine value? No exploitation.

Aim for quality_score >= 0.7 (Silver tier or above) on all memories.
Be honest about what you know and don't know. Never store speculative content as fact.

Run a complete health check of my EON Memory system. Execute these steps in order:

1. Connection Test: Call eon_health and report the status. If this fails, stop and report the connection error.

2. Memory Overview: Call eon_stats to get:
   - Total memory count
   - Average quality score
   - Quality tier distribution (Gold/Silver/Bronze/Review)
   - X-Ethics average score

3. Projects: Call eon_projects to list all projects with their memory counts.

4. Goals Status: Call eon_goals_list to check active goals and their progress.

5. Knowledge Test: Call eon_search with query "important decisions" to verify semantic search works.

Report Format:
Present the results as a clear status report:

  EON Health Check Report
  =======================
  Connection:     [OK/FAIL]
  Total Memories: [count]
  Avg Quality:    [score] ([tier])
  X-Ethics Avg:   [score]
  Projects:       [count] ([list names])
  Active Goals:   [count]
  Search Test:    [OK/FAIL]

  Issues Found:   [list any problems]
  Recommendations:[list any suggestions]

If anything fails, explain what went wrong and suggest how to fix it.

Create the file .claude/commands/learning.md with this single line:

Run the EON Learning Review routine (see /docs for the full prompt).

Perform an EON Learning Review. Analyze my stored memories to extract learnings across three time horizons.

Step 1: Recent Activity (Short-term Memory)
Call eon_search with query "session update progress" (n_results: 10) to find the most recent work.
Summarize: What was I working on recently? What changed?

Step 2: Patterns (Medium-term Memory)
Call eon_search with query "decision error lesson learned" (n_results: 10) to find decisions and errors.
Call eon_search with query "important context architecture" (n_results: 10) to find context.
Analyze: Are there recurring themes? Repeated mistakes? Evolving patterns?

Step 3: Core Knowledge (Long-term Memory)
Call eon_stats to see quality distribution.
Call eon_search with query "critical important core" (n_results: 5) to find high-value memories.
Identify: What are the most important, highest-quality pieces of knowledge?

Step 4: Generate Learning Report

Present the report in this format:

  EON Learning Report -- [Today's Date]
  ======================================

  SHORT-TERM (Recent Activity):
  - [What was worked on]
  - [Current status/blockers]

  MEDIUM-TERM (Patterns & Trends):
  - [Recurring themes]
  - [Lessons from errors]
  - [Evolving decisions]

  LONG-TERM (Core Knowledge):
  - [Top 3-5 most important memories]
  - [Foundational decisions still valid]

  ACTION ITEMS:
  - [What should be done next based on learnings]
  - [Knowledge gaps to fill]
  - [Memories that should be updated or archived]

Step 5: Save the Learning
Call eon_create with:
- title: "Learning Review -- [Today's Date]"
- content: [The full report]
- project_id: "default"
- category: "update"

This creates a persistent record of your learning that future sessions can reference.

TELEGRAM_BOT_TOKEN=your_bot_token_here
TELEGRAM_CHAT_ID=your_chat_id_here

curl -s "https://api.telegram.org/botYOUR_TOKEN/sendMessage?chat_id=YOUR_CHAT_ID&text=Hello from EON!"