MCP Server Template
A template MCP server implementing the 2026-07-28 spec with dynamic tool loading, API key authentication, and sample calculator and timestamp tools.
README
MCP Server Template with Modular Tools
A Node.js implementation of an MCP (Model Context Protocol) server built on the 2026-07-28 specification: stateless, per-request metadata, Streamable HTTP.
Features
- MCP 2026-07-28 only: stateless core,
server/discover,subscriptions/listen,resultType, cacheable list results, request-metadata headers — no handshake-era code paths to maintain - Dynamic Tool Loading: automatically discovers and loads tools from
/tools - Typed tool results:
outputSchema+structuredContent, input validation - API Key Authentication with an RFC 9728
WWW-Authenticatechallenge - Sample Tools: calculator and timestamp for demonstration
What changed in 2026-07-28
The 2026-07-28 revision made MCP a stateless request/response protocol. This
server implements that revision and nothing older — a client on 2025-11-25 or
earlier gets a 400 telling it which version to use. If you are pointing an
existing client at this server, these are the changes that matter:
| Removed | Replacement |
|---|---|
initialize / notifications/initialized |
Per-request _meta on every request |
Mcp-Session-Id header, DELETE /mcp |
No protocol sessions — pass explicit handles as tool arguments |
GET /mcp SSE stream, resources/subscribe |
subscriptions/listen (one long-lived POST-response stream) |
ping, logging/setLevel, notifications/roots/list_changed |
Removed; log level is per-request via _meta |
Last-Event-ID resumability, SSE event ids |
None — re-issue the request with a new id |
Server-initiated sampling/createMessage, elicitation/create, roots/list |
Multi Round-Trip Requests (resultType: "input_required") |
| JSON-RPC batching | One JSON-RPC message per POST |
Added: server/discover, the required resultType field on every result,
ttlMs/cacheScope on list results, required MCP-Protocol-Version /
Mcp-Method / Mcp-Name headers, x-mcp-header tool parameters, and the
-32020/-32021/-32022 error codes. Roots, Sampling and Logging are now
deprecated and are not advertised by this server.
Quick Start
- Install dependencies:
npm install
- Set up environment:
cp .env.example .env
- Start the server:
npm start
- Run the conformance tests:
npm test
Server runs on http://127.0.0.1:3202.
🔐 Authentication
An API key is required for all requests (except /health):
curl -H "Authorization: Bearer your-api-key" http://127.0.0.1:3202/health
X-API-Key and an ?api_key= query parameter also work, but the Authorization
header is what MCP clients send. A 401 carries a WWW-Authenticate: Bearer
challenge. Set MCP_AUTHORIZATION_SERVER to advertise a real OAuth 2.0
authorization server — the challenge then points at
/.well-known/oauth-protected-resource (RFC 9728), which is what MCP clients
probe. Client credentials must be keyed by issuer, and new clients should prefer
Client ID Metadata Documents over Dynamic Client Registration, which this
revision deprecates.
📡 Talking to the server
Every request carries its protocol version, client identity and capabilities in
params._meta, and mirrors method (and name/uri) into HTTP headers. A
mismatch between headers and body is rejected with 400 and error -32020.
Discovery
curl -X POST http://127.0.0.1:3202/mcp -H "Authorization: Bearer your-api-key" -H "Content-Type: application/json" -H "Accept: application/json, text/event-stream" -H "MCP-Protocol-Version: 2026-07-28" -H "Mcp-Method: server/discover" -d '{"jsonrpc":"2.0","id":1,"method":"server/discover","params":{"_meta":{"io.modelcontextprotocol/protocolVersion":"2026-07-28","io.modelcontextprotocol/clientInfo":{"name":"curl","version":"1.0.0"},"io.modelcontextprotocol/clientCapabilities":{}}}}'
Calling a tool
tools/call additionally requires the Mcp-Name header, matching params.name:
curl -X POST http://127.0.0.1:3202/mcp -H "Authorization: Bearer your-api-key" -H "Content-Type: application/json" -H "Accept: application/json, text/event-stream" -H "MCP-Protocol-Version: 2026-07-28" -H "Mcp-Method: tools/call" -H "Mcp-Name: calculator" -d '{"jsonrpc":"2.0","id":2,"method":"tools/call","params":{"name":"calculator","arguments":{"operation":"multiply","operand1":7,"operand2":8},"_meta":{"io.modelcontextprotocol/protocolVersion":"2026-07-28","io.modelcontextprotocol/clientCapabilities":{}}}}'
Response:
{
"jsonrpc": "2.0",
"id": 2,
"result": {
"resultType": "complete",
"content": [{ "type": "text", "text": "{\"operation\":\"multiply\",\"operands\":[7,8],\"result\":56}" }],
"structuredContent": { "operation": "multiply", "operands": [7, 8], "result": 56 },
"_meta": { "io.modelcontextprotocol/serverInfo": { "name": "mcp-server", "version": "2.0.0" } }
}
}
Add "progressToken" to _meta and send Accept: text/event-stream to get the
response as a stream with notifications/progress ahead of the result.
Change notifications
curl -N -X POST http://127.0.0.1:3202/mcp -H "Authorization: Bearer your-api-key" -H "Content-Type: application/json" -H "Accept: text/event-stream" -H "MCP-Protocol-Version: 2026-07-28" -H "Mcp-Method: subscriptions/listen" -d '{"jsonrpc":"2.0","id":1,"method":"subscriptions/listen","params":{"notifications":{"toolsListChanged":true},"_meta":{"io.modelcontextprotocol/protocolVersion":"2026-07-28","io.modelcontextprotocol/clientCapabilities":{}}}}'
The first message is notifications/subscriptions/acknowledged, echoing the
filters the server will honour. Every message on the stream is tagged with
io.modelcontextprotocol/subscriptionId. Run with MCP_WATCH_TOOLS=true and
edit a file in tools/ to see notifications/tools/list_changed arrive.
Closing the stream is the cancellation signal — there is no DELETE.
🛠️ Available Tools
| Tool | Description |
|---|---|
calculator |
Add, subtract, multiply, divide — returns structuredContent |
timestamp |
Current time as ISO 8601, Unix epoch, or human-readable |
⚙️ Creating Your Tools
1. Tool File Structure
Create tools/your-tool.js:
const TOOL_DEFINITION = {
name: "your_tool",
title: "Your Tool",
description: "What your tool does",
inputSchema: {
type: "object",
properties: {
param1: { type: "string", description: "Parameter description" }
},
required: ["param1"],
additionalProperties: false
},
// Optional but recommended: lets clients validate structuredContent.
outputSchema: {
type: "object",
properties: { result: { type: "string" } },
required: ["result"]
}
};
async function execute(args = {}, context = {}) {
const { param1 } = args;
// context.reportProgress({ progress, total, message }) streams progress
// when the client sent a progressToken.
// context.signal aborts when the client closes the stream.
// context.clientInfo / context.clientCapabilities describe the caller.
const structuredContent = { result: `Processed: ${param1}` };
return {
content: [{ type: "text", text: JSON.stringify(structuredContent) }],
structuredContent
};
}
module.exports = { definition: TOOL_DEFINITION, execute };
Arguments are validated against inputSchema before execute runs. Throwing
from execute produces a tool execution error (isError: true) rather than a
JSON-RPC error, so the model can self-correct.
2. Auto-Loading
Save the file in /tools and restart the server (or set MCP_WATCH_TOOLS=true
to hot-reload and notify subscribers).
3. Stateful tools
MCP has no protocol-level session. If a tool needs state across calls, return an opaque handle and accept it as an argument on later calls — document its lifetime in the tool description so the model knows when to create a new one.
4. Asking the client for input (MRTR)
Instead of sending a server-initiated elicitation/create request, return an
input-required result and let the client retry:
return {
resultType: "input_required",
inputRequests: {
github_login: {
method: "elicitation/create",
params: {
mode: "form",
message: "Please provide your GitHub username",
requestedSchema: {
type: "object",
properties: { name: { type: "string" } },
required: ["name"]
}
}
}
},
// Anything you need to resume; it comes back on the retry.
requestState: "..."
};
The retry arrives as a new tools/call with context.inputResponses and
context.requestState populated.
5. Exposing a parameter as an HTTP header
Annotate a primitive, statically reachable property with x-mcp-header so
intermediaries can route on it without parsing the body:
region: { type: "string", description: "...", "x-mcp-header": "Region" }
Conforming clients then send Mcp-Param-Region: us-west1, and the server
rejects any request where the header and the argument disagree. Never annotate
secrets — header values are visible to every intermediary on the path.
🔗 MCP Client Integration
{
"mcpServers": {
"template": {
"type": "http",
"url": "http://127.0.0.1:3202/mcp",
"headers": {
"Authorization": "Bearer your-api-key"
}
}
}
}
📁 Layout
mcp-server.js Express app, routing, request validation, dispatch
lib/protocol.js Version constants, _meta keys, error codes, message builders
lib/headers.js Request-metadata headers, base64 sentinel, x-mcp-header
lib/schema.js Tool argument validation
lib/sse.js SSE response streams (no resumability, per spec)
lib/subscriptions.js subscriptions/listen stream management
lib/security.js Origin validation, API key auth, RFC 9728 metadata
tools/ Auto-loaded tools
test/ Protocol conformance tests
Endpoints
| Endpoint | Purpose |
|---|---|
POST /mcp |
The MCP endpoint — the only method it accepts |
GET /mcp, DELETE /mcp |
405: the standalone SSE stream and session termination are gone |
GET /health |
Status, protocol version, open subscriptions |
GET /tools/config |
Tool definitions for wiring into an agent config |
GET /.well-known/oauth-protected-resource |
RFC 9728 metadata, when configured |
License
This server is provided as-is for demonstration purposes. Please review and enhance security measures before production use.
推荐服务器
Baidu Map
百度地图核心API现已全面兼容MCP协议,是国内首家兼容MCP协议的地图服务商。
Playwright MCP Server
一个模型上下文协议服务器,它使大型语言模型能够通过结构化的可访问性快照与网页进行交互,而无需视觉模型或屏幕截图。
Magic Component Platform (MCP)
一个由人工智能驱动的工具,可以从自然语言描述生成现代化的用户界面组件,并与流行的集成开发环境(IDE)集成,从而简化用户界面开发流程。
Audiense Insights MCP Server
通过模型上下文协议启用与 Audiense Insights 账户的交互,从而促进营销洞察和受众数据的提取和分析,包括人口统计信息、行为和影响者互动。
VeyraX
一个单一的 MCP 工具,连接你所有喜爱的工具:Gmail、日历以及其他 40 多个工具。
graphlit-mcp-server
模型上下文协议 (MCP) 服务器实现了 MCP 客户端与 Graphlit 服务之间的集成。 除了网络爬取之外,还可以将任何内容(从 Slack 到 Gmail 再到播客订阅源)导入到 Graphlit 项目中,然后从 MCP 客户端检索相关内容。
Kagi MCP Server
一个 MCP 服务器,集成了 Kagi 搜索功能和 Claude AI,使 Claude 能够在回答需要最新信息的问题时执行实时网络搜索。
e2b-mcp-server
使用 MCP 通过 e2b 运行代码。
Neon MCP Server
用于与 Neon 管理 API 和数据库交互的 MCP 服务器
Exa MCP Server
模型上下文协议(MCP)服务器允许像 Claude 这样的 AI 助手使用 Exa AI 搜索 API 进行网络搜索。这种设置允许 AI 模型以安全和受控的方式获取实时的网络信息。