Petstore MCP Server
A read-only MCP server that exposes Swagger Petstore v2 GET endpoints as tools for searching available pets, fetching pet details, and checking inventory.
README
Theorvane Petstore MCP Example
A single runnable example showing a truthful end-to-end path from the public Swagger Petstore v2 demo API to a TypeChain @Agent() that uses an actual local MCP HTTP session.
What this proves
Swagger Petstore v2 GET API
→ injected PetstoreClient
→ TypeMCP @McpServer() / @McpTool()
→ TypeMCP Streamable HTTP handler
→ local 127.0.0.1 ephemeral-port MCP runtime
→ official MCP SDK client session
→ TypeChain @Agent() + @Tool() MCP-client façade
→ TypeChain buildAgent() / LangChain agent loop
The fixture run proves the MCP SDK client performs protocol initialization, tools/list, and tools/call over loopback Streamable HTTP. It also proves a LangChain FakeToolCallingModel selects search_available_pets, and the TypeChain @Agent() façade sends that call through the MCP client rather than calling the REST client or TypeMCP server directly.
Prerequisites
- Node.js 20 or newer
- npm 10 or newer
npm ci
npm run check
Run the deterministic MCP agent example
npm run example:petstore:mcp-agent:fixture
It starts an in-process Node HTTP server bound only to 127.0.0.1 on an ephemeral port, connects the official MCP SDK StreamableHTTPClientTransport, discovers the tools, runs the agent loop, and shuts down the MCP client session and HTTP server in finally blocks.
Expected result contains fixture data:
{
"discoveredTools": [
"search_available_pets",
"get_pet",
"get_petstore_inventory"
],
"tool": "search_available_pets",
"result": [
{ "id": 1, "name": "Milo", "status": "available" },
{ "id": 2, "name": "Nori", "status": "available" }
]
}
For the narrower protocol-only smoke run (loopback URL, 404 route boundary, tools/list, and tools/call), use:
npm run example:petstore:mcp:fixture
Optional public REST smoke run
npm run example:petstore:live
This calls the mutable public Swagger Petstore demo API. Its data and uptime are outside this repository's control, so it is not used by tests or CI. The command is a GET-only REST smoke run; it does not demonstrate MCP client connectivity or a model-backed agent.
Read-only Petstore contract
PetstoreClient has a fixed base URL, https://petstore.swagger.io/v2, and only sends GET requests. The TypeMCP server exposes exactly three tools:
| MCP tool | REST operation |
|---|---|
search_available_pets |
GET /pet/findByStatus?status=available |
get_pet |
GET /pet/{petId} |
get_petstore_inventory |
GET /store/inventory |
The example has no create, update, delete, API-key, credential, environment-configured API host, or public listener path.
Application-owned model runtime
The fixture uses FakeToolCallingModel, not a live provider. An application that wants a real model must install/configure that model itself, then pass it together with a connected MCP client to the agent builder:
import { buildPetstoreMcpAgent } from "./examples/typechain-petstore-mcp-agent.js";
// The consuming application chooses, authenticates, and authorizes this model.
declare const applicationModel: Parameters<
typeof buildPetstoreMcpAgent
>[0]["model"];
// The application owns the lifecycle of this client connection.
declare const connectedMcpClient: Parameters<
typeof buildPetstoreMcpAgent
>[0]["client"];
const agent = buildPetstoreMcpAgent({
model: applicationModel,
client: connectedMcpClient,
});
This repository deliberately does not select a provider/model, install provider SDKs, read credentials or process.env, run a live LLM command, or implement authorization/retry/audit/redaction policy.
Example-only runtime boundary
The loopback runtime is a deterministic integration example, not a production deployment recipe. It intentionally does not provide public hosting, authentication, authorization, durable sessions, a reverse proxy, observability, deployment configuration, or persistent MCP lifecycle management. A production application must own those decisions.
Development
npm run lint
npm run build
npm test
npm run audit:prod
npm run check
Packages
@theorvane/type-chain0.1.1@theorvane/type-mcp0.2.2or later within the0.2.xrange
License
MIT © Theorvane
Recommended Servers
playwright-mcp
A Model Context Protocol server that enables LLMs to interact with web pages through structured accessibility snapshots without requiring vision models or screenshots.
Magic Component Platform (MCP)
An AI-powered tool that generates modern UI components from natural language descriptions, integrating with popular IDEs to streamline UI development workflow.
Audiense Insights MCP Server
Enables interaction with Audiense Insights accounts via the Model Context Protocol, facilitating the extraction and analysis of marketing insights and audience data including demographics, behavior, and influencer engagement.
VeyraX MCP
Single MCP tool to connect all your favorite tools: Gmail, Calendar and 40 more.
graphlit-mcp-server
The Model Context Protocol (MCP) Server enables integration between MCP clients and the Graphlit service. Ingest anything from Slack to Gmail to podcast feeds, in addition to web crawling, into a Graphlit project - and then retrieve relevant contents from the MCP client.
Kagi MCP Server
An MCP server that integrates Kagi search capabilities with Claude AI, enabling Claude to perform real-time web searches when answering questions that require up-to-date information.
E2B
Using MCP to run code via e2b.
Neon Database
MCP server for interacting with Neon Management API and databases
Exa Search
A Model Context Protocol (MCP) server lets AI assistants like Claude use the Exa AI Search API for web searches. This setup allows AI models to get real-time web information in a safe and controlled way.
Qdrant Server
This repository is an example of how to create a MCP server for Qdrant, a vector search engine.