vreko-mcp-server
v3.1.1
Published
Vreko MCP Proxy - HTTP reverse proxy forwarding MCP requests to the API server
Readme
What is Vreko MCP Server?
Vreko MCP Server is a Model Context Protocol (MCP) server that gives AI coding agents a session memory and intelligence layer. It enables Claude, Cursor, Windsurf, and any MCP-compatible assistant to:
- Brief themselves before every task - past learnings, active warnings, risk context
- Check vitals mid-session - activity level, risk pressure, trajectory
- Capture discoveries - patterns, gotchas, and decisions that carry forward to future sessions
- Close with ceremony - session outcome recorded, compound intelligence grows
Key Capabilities:
| Feature | Description |
|---------|-------------|
| Session Intelligence | vreko → vreko_pulse → vreko_learn → vreko_end - the agentic coding loop |
| Pattern Memory | Learns your codebase’s specific failure modes over time |
| Multi-Assistant Support | Works with Claude, Cursor, Windsurf, and any MCP-compatible client |
| Cloud & Local Modes | Deploy to Fly.io or run locally via CLI |
| Security-First | API key validation, path traversal protection, CORS security |
Quick Start
1. Install
Option A: npx (No Installation)
npx vreko-mcp-serverOption B: Global Install
npm install -g vreko-mcp-server
vreko-mcp-serverOption C: Docker
docker run -p 8080:8080 -e VREKO_API_KEY=your_key vreko/mcp-server2. Configure Your AI Assistant
Claude Desktop Config (~/Library/Application Support/Claude/claude_desktop_config.json):
{
"mcpServers": {
"vreko": {
"command": "npx",
"args": ["vreko-mcp-server"],
"env": {
"VREKO_API_KEY": "YOUR_VREKO_API_KEY"
}
}
}
}Cursor Config (Settings → MCP):
{
"mcpServers": {
"vreko": {
"command": "npx",
"args": ["vreko-mcp-server"],
"env": {
"VREKO_API_KEY": "YOUR_VREKO_API_KEY"
}
}
}
}3. Verify Installation
# Health check
curl http://localhost:8080/health
# Expected response:
{"status":"healthy","version":"1.0.0","timestamp":"2026-01-..."}Architecture
Wraps the CLI-based MCP server (packages/mcp) in an Express HTTP server that can be deployed to Fly.io or other cloud providers. This allows Claude Desktop, Cursor, and other AI tools to connect to Vreko over HTTPS instead of requiring local installation.
┌─────────────────────────────────────────────────────────────┐
│ AI Assistant Clients │
│ ┌──────────────┐ ┌──────────────┐ ┌──────────────────┐ │
│ │Claude Desktop│ │ Cursor │ │ Other MCP Tools │ │
│ └──────┬───────┘ └──────┬───────┘ └────────┬─────────┘ │
└─────────┼─────────────────┼───────────────────┼────────────┘
│ │ │
└─────────────────┼───────────────────┘
▼ HTTPS
┌─────────────────────────────────────────────────────────────┐
│ apps/mcp-server │
│ ┌────────────────────────────────────────────────────────┐ │
│ │ Express HTTP Server • API Key Auth • CORS │ │
│ │ • Request Validation • Error Handling • Logging │ │
│ └────────────────────┬───────────────────────────────────┘ │
└───────────────────────┼─────────────────────────────────────┘
▼ Stdio
┌─────────────────────────────────────────────────────────────┐
│ packages/mcp │
│ ┌────────────────────────────────────────────────────────┐ │
│ │ MCP Protocol Handler • Tool Registry • Context │ │
│ └────────────────────┬───────────────────────────────────┘ │
└───────────────────────┼─────────────────────────────────────┘
▼
┌─────────────────────────────────────────────────────────────┐
│ Vreko Platform │
│ ┌──────────────┐ ┌──────────────┐ ┌──────────────────┐ │
│ │Pattern Memory│ │ Snapshot │ │ Intelligence │ │
│ │ Engine │ │ Engine │ │ Layer │ │
│ └──────────────┘ └──────────────┘ └──────────────────┘ │
└─────────────────────────────────────────────────────────────┘Component Responsibilities
| Component | Responsibility | |-----------|----------------| | HTTP Server | Handles HTTPS requests, authentication, request/response formatting | | MCP Protocol | Implements Model Context Protocol spec, tool registration, JSON-RPC | | Vreko Core | Pattern analysis, snapshot management, learning capture |
Local Development
Build
pnpm buildRun
# Development
PORT=8080 NODE_ENV=development node dist/index.js
# With API key
VREKO_API_KEY=YOUR_VREKO_API_KEY PORT=8080 node dist/index.jsTest Health
curl http://localhost:8080/healthTest MCP Call
# Initialize MCP session (standard JSON-RPC initialize handshake)
curl -X POST http://localhost:8080/mcp \
-H "Content-Type: application/json" \
-H "x-workspace-path: /path/to/your/project" \
-H "x-api-key: YOUR_VREKO_API_KEY" \
-d '{"jsonrpc":"2.0","id":1,"method":"initialize","params":{"protocolVersion":"2024-11-05","capabilities":{},"clientInfo":{"name":"test","version":"1.0"}}}'Deployment to Fly.io
Quick Deploy (Recommended)
From the monorepo root:
# Full deployment with pre-flight checks and verification
pnpm --filter=vreko-mcp-server deploy
# Dry run (validates without deploying)
pnpm --filter=vreko-mcp-server deploy:dry-runThe deploy script automatically:
- ✅ Validates Fly CLI is installed and authenticated
- ✅ Checks for uncommitted changes
- ✅ Builds and validates the bundle
- ✅ Deploys to Fly.io
- ✅ Verifies health and
/bridge/pushendpoints after deployment
Alternative Deploy Methods
# Quick deploy (skips pre-flight checks)
pnpm --filter=vreko-mcp-server deploy:quick
# Force immediate deployment (no rolling update)
pnpm --filter=vreko-mcp-server deploy:force
# Check deployment status
pnpm --filter=vreko-mcp-server deploy:status
# View live logs
pnpm --filter=vreko-mcp-server deploy:logs
# View releases for rollback
pnpm --filter=vreko-mcp-server deploy:rollbackFirst-Time Setup
If deploying for the first time:
# 1. Login to Fly.io
fly auth login
# 2. Create the app (only once)
fly apps create vreko-mcp
# 3. Set secrets
fly secrets set VREKO_API_KEY=YOUR_VREKO_API_KEY -a vreko-mcp
fly secrets set CORS_ORIGIN=https://your-domain.com -a vreko-mcp
# 4. Deploy
pnpm --filter=vreko-mcp-server deployCI/CD Deployment
Automatic deployments are triggered on push to main when changes are detected in:
apps/mcp-server/**packages/mcp/**,packages/core/**,packages/intelligence/**
Required GitHub Secret: FLY_API_TOKEN
Generate a token: fly tokens create deploy -x 999999h
Verify Deployment
# Health check
curl https://vreko-mcp.fly.dev/health
# Bridge endpoint test
curl -X POST https://vreko-mcp.fly.dev/bridge/push \
-H "Content-Type: application/json" \
-d '{"workspaceId":"ws_00000000000000000000000000000000","observations":[]}'Expected responses:
{"status":"healthy","uptime":123.456,"timestamp":"2025-01-06T...","version":"1.0.0"}
{"received":true,"observationsCount":0,"changesCount":0}Rollback
If something goes wrong:
# List recent releases
fly releases -a vreko-mcp
# Rollback to a previous version
fly releases rollback v123 -a vreko-mcpConfiguration
Environment Variables
| Variable | Default | Description |
|----------|---------|-------------|
| PORT | 8080 | Server port |
| NODE_ENV | development | Environment (production/development) |
| VREKO_API_KEY | (required) | API key for authentication |
| CORS_ORIGIN | * | Allowed CORS origins (comma-separated) |
| LOG_LEVEL | info | Logging level |
Fly.io Secrets
Set via fly secrets set KEY=value:
fly secrets set VREKO_API_KEY=YOUR_VREKO_API_KEY
fly secrets set CORS_ORIGIN=https://your-app.comView secrets:
fly secrets listThe V2 Agentic Workflow
Vreko MCP exposes a 4-tool session API designed for the agentic coding loop. Configure once, and your AI assistant uses it automatically:
| Tool | When | What it does |
|------|------|--------------|
| vreko | Task start | Returns intelligence briefing: recent learnings, active warnings, session lineage |
| vreko_pulse | Mid-session | Read-only vitals: pulse level, risk pressure, trajectory |
| vreko_learn | Any discovery | Records a pattern, gotcha, or decision for future sessions |
| vreko_end | Task complete | Closes session with ceremony - outcome, carry-forward context |
vreko({ task: "add rate limiting to API" })
→ briefing: 3 past learnings, 1 active warning, lineage from last session
[... agent works ...]
vreko_pulse()
→ pulse: racing, pressure: 71%, trajectory: increasing - snapshot recommended
vreko_learn({ insight: "rate limiting config is shared - test in staging first" })
vreko_end({ outcome: "completed" })
→ ceremony: 4 files modified, 1 learning captured, context carried forwardThe full consolidated surface is also available: check, advise, guide, safe_to_write, intelligence and refactoring tools. See MCP documentation →
API Reference
GET /health
Health check endpoint.
Response:
{
"status": "healthy",
"uptime": 123.456,
"timestamp": "2024-12-30T...",
"version": "1.0.0"
}POST /mcp
Main MCP endpoint - handles standard JSON-RPC 2.0 requests per the MCP protocol spec.
Headers:
x-workspace-path: /absolute/path/to/project
x-api-key: YOUR_VREKO_API_KEY
mcp-session-id: (returned from initialize, required for subsequent calls)Example - initialize session:
{"jsonrpc":"2.0","id":1,"method":"initialize","params":{"protocolVersion":"2024-11-05","capabilities":{},"clientInfo":{"name":"my-client","version":"1.0"}}}Response: Standard JSON-RPC 2.0 response. The mcp-session-id header in the response must be included in all subsequent requests.
Security
- Authentication: Required
apiKeyparameter (validated againstVREKO_API_KEY) - HTTPS Only: Fly.io auto-configures HTTPS on port 443
- CORS: Configurable via
CORS_ORIGINenvironment variable - Non-root user: Container runs as unprivileged user
- Request timeouts: 30-second timeout on MCP calls
- Rate limiting: Consider adding in production (not yet implemented)
Monitoring
Logs
fly logsMetrics
# CPU/Memory usage
fly scale show
# App status
fly statusScaling
Edit fly.toml:
[scaling]
min_count = 1
max_count = 5Then deploy:
fly deployTroubleshooting
"Connection refused"
Ensure Fly.io app is running:
fly status"Unauthorized"
Check that API key is set:
fly secrets listShould show VREKO_API_KEY.
"MCP server timeout"
Default timeout is 30 seconds. Check workspace size and API key validity.
Logs show errors
View detailed logs:
fly logs --followIntegration with VS Code Extension
Quick Setup (2 steps)
- Get your API key from deployment output (look for
YOUR_VREKO_API_KEY) - Configure VS Code settings:
{
"vreko.mcp.enabled": true,
"vreko.mcp.serverUrl": "https://vreko-mcp.fly.dev",
"vreko.mcp.authType": "apikey"
}- Store your API key securely using VS Code command palette:
- Press
Cmd+Shift+P/Ctrl+Shift+P - Run:
Vreko: Set MCP Auth Token (Secure) - Paste your API key:
YOUR_VREKO_API_KEY
- Press
Configuration Options
| Setting | Default | Description |
|---------|---------|-------------|
| vreko.mcp.enabled | true | Enable MCP integration |
| vreko.mcp.serverUrl | https://mcp.vreko.dev | Remote MCP server URL |
| vreko.mcp.authType | bearer | Authentication type (bearer or apikey) |
| vreko.mcp.timeout | 5000 | Request timeout in milliseconds |
Programmatic Usage
The VS Code extension automatically uses RemoteMCPClient when serverUrl is configured:
const client = new RemoteMCPClient({
serverUrl: 'https://vreko-mcp.fly.dev',
apiKey: 'YOUR_VREKO_API_KEY',
authType: 'apikey',
timeout: 5000,
maxRetries: 3,
});
await client.connect();See apps/vscode/src/services/RemoteMCPClient.ts for implementation details.
Health Check
Verify the server is accessible:
curl https://vreko-mcp.fly.dev/healthExpected response:
{"status":"healthy","uptime":123.45,"timestamp":"2025-12-31T...","version":"1.0.0"}Cost
Fly.io free tier includes:
- 3 shared-cpu-1x VMs (256MB each)
- 160GB bandwidth/month
This MCP server easily fits within free tier for individual/small team use.
Status
| Metric | Value | |--------|-------| | Version | 3.1.0 | | License | Apache-2.0 | | Status | Production Ready | | Node.js | >= 20.0.0 |
