Apollo MCP Demo
This MCP server exposes approved GraphQL operations from an Apollo GraphQL API as AI-accessible tools. It sits between MCP clients and the GraphQL API, translating tool calls into GraphQL queries and returning results in MCP format.
README
Apollo MCP Demo
This folder sets up the same core architecture Apollo demonstrates for MCP:
- a normal GraphQL API,
- an Apollo MCP Server in front of it,
- a small allowlist of GraphQL operations that become MCP tools.
This version uses the standalone Apollo MCP Server container instead of the rover init --mcp wizard because it is simpler to run locally without GraphOS credentials.
What gets started
api: a tiny Apollo Server onhttp://localhost:4001/mcp: Apollo MCP Server onhttp://localhost:8000/mcp
Architecture Diagram
flowchart LR
subgraph LocalMachine["Local machine"]
Client["MCP client\nInspector / Codex / Cursor"]
subgraph Docker["docker compose"]
MCP["Apollo MCP Server\n:8000/mcp"]
API["Apollo GraphQL API\n:4001"]
end
Ops["graphql/operations/*.graphql\nApproved operations"]
Schema["graphql/api.graphql\nSchema snapshot"]
end
Client -->|"MCP over Streamable HTTP"| MCP
Ops -->|"tool definitions"| MCP
Schema -->|"schema metadata"| MCP
MCP -->|"GraphQL queries/mutations"| API
API -->|"GraphQL JSON response"| MCP
MCP -->|"MCP tool results"| Client
Run it
cd C:\Users\muham\Documents\apollo-mcp
npm install
docker compose up --build
In a second terminal, inspect the MCP server:
npx @modelcontextprotocol/inspector
Then open the Inspector URL it prints, choose:
- Transport Type:
Streamable HTTP - URL:
http://127.0.0.1:8000/mcp
Click Connect, then List Tools.
You should see MCP tools derived from:
GetBooksGetBookByIdGetBooksByGenre
Optional client config
For MCP clients that use mcp-remote, the local config shape is:
{
"mcpServers": {
"apollo-demo": {
"command": "npx",
"args": [
"mcp-remote",
"http://127.0.0.1:8000/mcp"
]
}
}
}
What's actually happening
The GraphQL API is still the real data service. It knows the schema, resolves fields, and returns JSON for GraphQL queries.
Apollo MCP Server does not replace that API. It sits in front of it and translates approved GraphQL operations into MCP tools that an AI client can discover and call.
The important files are:
- mcp-config.yaml: tells Apollo MCP Server where the GraphQL endpoint lives, where the schema file is, where the allowed operations live, and how to expose MCP over HTTP.
- graphql/api.graphql: schema snapshot used by Apollo MCP Server so it can describe tools correctly.
- graphql/operations/GetBooks.graphql: one approved query that becomes one MCP tool.
- src/server.mjs: the actual GraphQL API implementation.
- docker-compose.yml: runs both services together on one local network.
The request flow is:
- Your MCP client connects to
http://localhost:8000/mcp. - Apollo MCP Server advertises tools based on the
.graphqloperation files. - The client calls one of those tools with arguments.
- Apollo MCP Server converts that tool call back into the underlying GraphQL operation.
- Apollo MCP Server sends that GraphQL request to
http://api:4001/. - The GraphQL API executes the query and returns data.
- Apollo MCP Server returns the result to the MCP client in MCP format.
Request Flow Diagram
sequenceDiagram
autonumber
participant C as MCP Client
participant M as Apollo MCP Server
participant O as Operation Files
participant G as GraphQL API
C->>M: Connect to http://localhost:8000/mcp
M->>O: Load approved .graphql operations
O-->>M: GetBooks / GetBookById / GetBooksByGenre
M-->>C: Advertise MCP tools
C->>M: Invoke tool with arguments
M->>M: Convert tool call to GraphQL operation
M->>G: Send GraphQL request to http://api:4001/
G-->>M: Return GraphQL JSON data
M-->>C: Return MCP-formatted tool result
Why there is no router here
Apollo Router is useful when:
- you are serving a federated supergraph,
- you want Apollo's GraphOS-managed runtime,
- or you want GraphQL and MCP bundled in Apollo Runtime.
For a simple local demo with one existing GraphQL API, Apollo's own deployment docs support using the standalone Apollo MCP Server container directly against that endpoint. That is what this project uses.
Sources
- Apollo MCP Server quickstart: https://www.apollographql.com/docs/apollo-mcp-server/quickstart
- Apollo MCP Server config reference: https://www.apollographql.com/docs/apollo-mcp-server/command-reference
- Apollo MCP Server deploy docs: https://www.apollographql.com/docs/apollo-mcp-server/deploy
- Apollo MCP Server debugging docs: https://www.apollographql.com/docs/apollo-mcp-server/debugging
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。