repaera

mcp-server

Community repaera
Updated

mcp-server

MCP server for integration with a Next.js AI chat.Built with Hono + Node.js, using Streamable HTTP transport (MCP spec 2025-03-26).

Target Adapter Persistence
Render / Fly.io InMemoryAdapter ❌ lost on restart
Cloudflare Workers D1Adapter ✅ persisted in D1 SQLite

Quickstart (Node.js / local)

cd mcp-server
npm install
cp .env.example .env   # leave everything empty for dev

npm run dev
# → http://localhost:3001/mcp

Set this in Next.js .env.local:

MCP_URL=http://localhost:3001/mcp

Structure

mcp-server/
├── src/
│   ├── index.ts        # Node.js entry point (Render / Fly.io) — InMemoryAdapter
│   ├── worker.ts       # Cloudflare Workers entry point — D1Adapter
│   ├── auth.ts         # Extract userId from Bearer + JWT headers
│   ├── tools.ts        # 5 MCP tools (list, add, complete, update, delete)
│   └── db/
│       ├── adapter.ts  # TodoDB interface
│       ├── memory.ts   # InMemoryAdapter
│       └── d1.ts       # D1Adapter (Cloudflare Workers only)
├── migrations/
│   └── 0001_init.sql   # D1 schema
├── wrangler.toml       # Cloudflare config
├── Dockerfile          # For Render / Fly.io
├── fly.toml
└── render.yaml

Tools

Tool Description Input
list_todos Display all todos filter?: "all"|"done"|"pending"
add_todo Add a new todo title: string
complete_todo Mark as completed id: string
update_todo Change the title id: string, title: string
delete_todo Delete a todo id: string

Auth

Two layers, both optional — leave empty for dev (fallback X-User-Id):

# .env
MCP_TOKEN=        # Bearer token — verifies service identity
MCP_JWT_SECRET=   # JWT secret — must match Next.js

TypeScript

This project requires "module": "NodeNext" and "moduleResolution": "NodeNext" in tsconfig.json. The MCP SDK exposes McpServer via a package.json exports wildcard — older settings like "moduleResolution": "bundler" with "module": "ESNext" resolve imports in CJS mode and silently skip the exports map, causing:

Cannot find module '@modelcontextprotocol/sdk/server/mcp.js'

NodeNext reads the project package.json "type": "module" field, resolves in ESM mode, and correctly follows the exports wildcard.

Developer Guide

Add a new tool

Edit src/tools.ts, add server.registerTool(...) inside the registerTools function.The tool will automatically be available in all adapters because db: TodoDB is passed from outside.

Manual testing with curl

The MCP Streamable HTTP transport (spec 2025-03-26) requires every POST request to include:

Accept: application/json, text/event-stream

The server needs to know the client can handle either a direct JSON response or an SSE stream — omitting this header returns a -32000 Not Acceptable error.

When MCP_TOKEN and MCP_JWT_SECRET are unset, auth falls back to the plain X-User-Id header.

# Health check
curl http://localhost:3001/

# List tools
curl -s -X POST http://localhost:3001/mcp \
  -H "Content-Type: application/json" \
  -H "Accept: application/json, text/event-stream" \
  -H "X-User-Id: dev-user" \
  -d '{"jsonrpc":"2.0","id":1,"method":"tools/list"}' | jq

# list_todos (all)
curl -s -X POST http://localhost:3001/mcp \
  -H "Content-Type: application/json" \
  -H "Accept: application/json, text/event-stream" \
  -H "X-User-Id: dev-user" \
  -d '{"jsonrpc":"2.0","id":2,"method":"tools/call","params":{"name":"list_todos","arguments":{}}}' | jq

# list_todos (filter: pending | done)
curl -s -X POST http://localhost:3001/mcp \
  -H "Content-Type: application/json" \
  -H "Accept: application/json, text/event-stream" \
  -H "X-User-Id: dev-user" \
  -d '{"jsonrpc":"2.0","id":3,"method":"tools/call","params":{"name":"list_todos","arguments":{"filter":"pending"}}}' | jq

# add_todo
curl -s -X POST http://localhost:3001/mcp \
  -H "Content-Type: application/json" \
  -H "Accept: application/json, text/event-stream" \
  -H "X-User-Id: dev-user" \
  -d '{"jsonrpc":"2.0","id":4,"method":"tools/call","params":{"name":"add_todo","arguments":{"title":"Learn MCP"}}}' | jq

# complete_todo
curl -s -X POST http://localhost:3001/mcp \
  -H "Content-Type: application/json" \
  -H "Accept: application/json, text/event-stream" \
  -H "X-User-Id: dev-user" \
  -d '{"jsonrpc":"2.0","id":5,"method":"tools/call","params":{"name":"complete_todo","arguments":{"id":"1"}}}' | jq

# update_todo
curl -s -X POST http://localhost:3001/mcp \
  -H "Content-Type: application/json" \
  -H "Accept: application/json, text/event-stream" \
  -H "X-User-Id: dev-user" \
  -d '{"jsonrpc":"2.0","id":6,"method":"tools/call","params":{"name":"update_todo","arguments":{"id":"1","title":"Review PR"}}}' | jq

# delete_todo
curl -s -X POST http://localhost:3001/mcp \
  -H "Content-Type: application/json" \
  -H "Accept: application/json, text/event-stream" \
  -H "X-User-Id: dev-user" \
  -d '{"jsonrpc":"2.0","id":7,"method":"tools/call","params":{"name":"delete_todo","arguments":{"id":"1"}}}' | jq
Auth header combinations
MCP_TOKEN set MCP_JWT_SECRET set Required headers
X-User-Id: <any> (omit → defaults to "dev-user")
Authorization: Bearer <token> + X-User-Id: <any>
X-User-Token: <jwt> (sub claim = userId)
Authorization: Bearer <token> + X-User-Token: <jwt>

Or use MCP Inspector:

npx @modelcontextprotocol/inspector http://localhost:3001/mcp

Add a D1 migration

# Create a new file in migrations/
echo "ALTER TABLE todos ADD COLUMN priority INTEGER DEFAULT 0;" \
  > migrations/0002_add_priority.sql

# Apply locally
npm run cf:migrate:local

# Apply to production
npm run cf:migrate

Deployment

Render

  1. Push the repo to GitHub (make sure mcp-server/render.yaml exists)
  2. Render dashboard → New → Blueprint → connect the repo
  3. Set env vars: MCP_TOKEN, MCP_JWT_SECRET
  4. Deploy

MCP URL:https://mcp-server.onrender.com/mcp

⚠️ Free tier sleeps after 15 minutes of idle time — the first request may take ~30 seconds to wake up.

Fly.io

# Install flyctl: https://fly.io/docs/hands-on/install-flyctl/
fly auth login

cd mcp-server

# First deployment — reads fly.toml and creates the app
fly launch --no-deploy

# Set secrets
fly secrets set MCP_TOKEN=$(openssl rand -hex 32)
fly secrets set MCP_JWT_SECRET=the-same-value-as-nextjs

# Deploy
fly deploy

Update after changes:

fly deploy

MCP URL:https://mcp-server.fly.dev/mcp

Cloudflare Workers + D1

1. Initial setup
npm install
npx wrangler login
2. Create a D1 database
npx wrangler d1 create todos-db

The output will display a database_id. Copy and paste it into wrangler.toml:

[[d1_databases]]
binding       = "TODOS_DB"
database_name = "todos-db"
database_id   = "PASTE_ID_HERE"
3. Run migrations
# Local dev
npm run cf:migrate:local

# Production
npm run cf:migrate
4. Set secrets
npx wrangler secret put MCP_TOKEN
# → enter the value when prompted

npx wrangler secret put MCP_JWT_SECRET
# → enter the SAME value as Next.js MCP_JWT_SECRET
5. Local development with D1
npm run cf:dev
# → http://localhost:8787/mcp (using local D1 via .wrangler/)
6. Deploy
npm run cf:deploy

MCP URL:https://mcp-server.your-subdomain.workers.dev/mcp

One-click deploy to Cloudflare

Note: Cloudflare one-click deploy cannot automatically set up D1 — you still need to perform steps 2–4 manually after deployment.

Next.js Integration

# .env.local

# Choose one:
MCP_URL=http://localhost:3001/mcp
MCP_URL=https://mcp-server.onrender.com/mcp
MCP_URL=https://mcp-server.fly.dev/mcp
MCP_URL=https://mcp-server.your-subdomain.workers.dev/mcp

# Auth — must match the MCP server
MCP_TOKEN=your-mcp-token
MCP_JWT_SECRET=your-jwt-secret

Update to — SSE → Streamable HTTP

This server uses Streamable HTTP. Update the transport:

// Before
transport: { type: "sse", url, headers }

// After
transport: { type: "http", url, headers }

Testing checklist

  • curl http://localhost:3001{"status":"ok","adapter":"memory"}
  • Chat: “Show all todos” → calls list_todos
  • Chat: “Add todo: learn MCP” → calls add_todo
  • Chat: “Mark todo #1 as done” → calls complete_todo
  • Chat: “Change todo #1 to: review PR” → calls update_todo
  • Chat: “Delete todo #1” → calls delete_todo
  • Check Sentry — no mcp:connect or mcp:close errors

MCP Server · Populars

MCP Server · New

    Dave-London

    Pare

    Dev tools, optimized for agents. Structured, token-efficient MCP servers for git, test runners, npm, Docker, and more.

    Community Dave-London
    luckyPipewrench

    Pipelock

    Firewall for AI agents. DLP scanning, SSRF protection, bidirectional MCP scanning, tool poisoning detection, and workspace integrity monitoring.

    Community luckyPipewrench
    longbridge

    Longbridge OpenAPI SDK

    LongPort OpenAPI SDK Base.

    Community longbridge
    longbridge

    Longbridge MCP

    LongPort OpenAPI SDK Base.

    Community longbridge
    ArcadeData

    arcadedb

    ArcadeDB Multi-Model Database, one DBMS that supports SQL, Cypher, Gremlin, HTTP/JSON, MongoDB and Redis. ArcadeDB is a conceptual fork of OrientDB, the first Multi-Model DBMS. ArcadeDB supports Vector Embeddings.

    Community ArcadeData