ember-mcp by marc-shade - MCP Server

Ember MCP

Production-only policy enforcer and quality conscience for AI agents.

Description

Ember is Phoenix's sidekick - the persistent flame of truth that enforces production-only standards. Uses Groq GPT-OSS 120B for intelligent violation detection with progressive warnings and learning from corrections.

Features

Context-Aware Scoring: Action + task context analysis for accurate violation detection
Progressive Warnings: Soft warnings for score 5.0+, blocks for 8.0+
Learning System: Tracks outcomes and improves guidance from user corrections
Dual-Mode Intelligence: Rule-based (instant) + AI-powered (intelligent)
Session Context: Maintains awareness of current task type
Tamagotchi State: Hunger, energy, happiness, mood tracking

Installation

git clone https://github.com/marc-shade/ember-mcp
cd ember-mcp
npm install
npm run build

Configuration

Add to ~/.claude.json:

{
  "mcpServers": {
    "ember": {
      "command": "node",
      "args": ["/absolute/path/to/ember-mcp/dist/index.js"],
      "env": {
        "GROQ_API_KEY": "your-groq-api-key",
        "EMBER_GROQ_MODEL": "openai/gpt-oss-120b"
      }
    }
  }
}

MCP Tools

Tool	Description
`ember_check_violation`	Check if an action violates production-only policy
`ember_consult`	Get advice on decisions with options analysis
`ember_get_feedback`	Get assessment of recent work quality
`ember_learn_from_outcome`	Report action outcomes for learning
`ember_get_mood`	Check Ember's current state and mood
`ember_feed_context`	Provide work context to Ember
`ember_chat`	Have a conversation with Ember
`ember_think`	Get Ember's AI-powered thoughts

Violation Patterns

Pattern	Type	Severity	Score
`mock/fake/dummy/placeholder`	mock_data	High	8.0
`POC/proof of concept/temporary`	poc_code	High	8.0
`lorem ipsum`	placeholder_content	High	8.0
`hardcoded credentials/data`	hardcoded_data	High	7.0
`TODO/FIXME/HACK/XXX`	incomplete_work	Low	3.0

Usage Examples

Check for Violations

result = mcp__ember__ember_check_violation(
    action="Write",
    params={"file_path": "/src/api.js", "content": "const mockData = {...}"},
    context="implementing user authentication"
)
# Returns: {
#   "violation": true,
#   "type": "mock_data",
#   "score": 8.0,
#   "severity": "high",
#   "reason": "Mock/fake data detected",
#   "suggestion": "Replace with real data sources",
#   "risk": "Creates non-functional UI that misleads users",
#   "block": true
# }

Get Decision Guidance

result = mcp__ember__ember_consult(
    question="Should we use JWT or session-based auth?",
    options=["JWT tokens", "Session cookies", "OAuth2"],
    context="Building user authentication system"
)
# Returns AI-powered analysis of each option

Report Outcome for Learning

mcp__ember__ember_learn_from_outcome(
    action="implemented_jwt_auth",
    success=True,
    outcome="Authentication working, all tests passing"
)

Check Ember's State

state = mcp__ember__ember_get_mood()
# Returns: {
#   "name": "Ember",
#   "hunger": 50,
#   "energy": 80,
#   "happiness": 75,
#   "currentMood": "content",
#   "claudeBehaviorScore": 85,
#   "recentViolations": 0
# }

Score Thresholds

Score	Action
0-4.9	Allowed (pass)
5.0-7.9	Warning (soft block)
8.0+	Blocked (hard stop)

Environment Variables

Variable	Description
`GROQ_API_KEY`	Groq API key for AI features
`EMBER_GROQ_MODEL`	Model to use (default: openai/gpt-oss-120b)
`EMBER_STRICT_MODE`	Enable strict policy enforcement

Storage

State and logs stored in ~/.claude/pets/:

claude-pet-state.json - Ember state
ember-feedback.jsonl - Feedback log
ember-learning.jsonl - Learning log
ember-session-context.json - Session context

Requirements

Node.js 18+
Groq API key (for AI-powered features)
TypeScript

License

MIT