Internal APIs
Let agents call the service you already run
Skip writing an MCP server by hand. Connect Cursor or Claude to your internal REST API.
Give agents access to your API.
Upload an OpenAPI file, pick the tools you want, and download a local MCP server. Then Cursor, Claude, or any other agent can call the APIs you already have.
Free to try. We don't keep your file after the session.
For developers connecting agents to real APIs
- Runs on your machine
- You choose the tools
- Your keys stay with you
- JSON and YAML
From OpenAPI to a running server
Three steps. No setup from scratch. You decide which endpoints become tools.
orders-api.yaml
Drop OpenAPI here
01Upload your OpenAPI
Drop in a JSON or YAML file. We read the paths, parameters, and response schemas.
- list_orderscreate_orderget_customer02
Keep only what you need
We suggest a tool for each endpoint. Turn off anything you don't want before you generate.
$ npx flashmcp-server
stdio · 12 tools ready
03Download and connect
Download a ready-to-run MCP server. Point Cursor, Claude, or any MCP host at it.
Review your tools first
Include or skip any endpoint
orders-api.yaml · 5 endpoints scanned
| Endpoint | Tool name | Type | Status |
|---|---|---|---|
| GET/v1/orders | list_orders | Tool | Included |
| POST/v1/orders | create_order | Tool | Included |
| GET/v1/customers/{id} | get_customer | Tool | Included |
| GET/v1/schema | order_schema | Resource | Included |
| DELETE/v1/orders/{id} | delete_order | Tool | Excluded |
You have the docs. You need the MCP server.
FlashMCP fills the gap when your OpenAPI file exists, but your agents still cannot use the API.
Prototyping
Try an agent idea the same day
Spend your time on the idea, not on scaffolding and wiring.
Existing docs
Turn OpenAPI into usable tools
If you already have the docs, FlashMCP turns them into tools your agents can call.
Teams
Share one server everyone can run
The output is plain files you can commit, review, and run the same way on every machine.
posts-api-mcp
Files
- mcp.ts
- .env.example
- package.json
- README.md
1import { McpServer } from "@modelcontextprotocol/sdk";2import { z } from "zod";3 4const server = new McpServer({5 name: "posts-api",6 version: "1.0.0",7});8 9server.registerTool(10 "create_post",11 {12 description: "Create a post",13 inputSchema: {14 username: z.string(),15 content: z.string(),16 },17 },18 async ({ username, content }) => {19 // POST /posts20 },21);What you get
A local MCP server you own
We generate the server file and the setup steps. You run it on your machine, and your credentials never leave your environment.
1. Download your server
Get an mcp.ts file with tools based on the endpoints you kept.
2. Follow the setup steps
Create a package.json, install the dependencies, and put the file in place.
3. Connect your agent
Add your API credentials on your machine, then point Cursor, Claude, or any MCP host at the server.
Before you upload
Quick answers on hosting, privacy, and what you get.
An MCP server lets AI agents call tools in a shared format. Once yours is running, an agent can use your API through it without custom integration code.
OpenAPI files in JSON or YAML. Support for older Swagger specs is on the roadmap.
No. FlashMCP gives you a project to run yourself. Requests to your API leave from your machine, and your credentials never pass through us.
We only use it to generate your server, and we don't keep it after the session ends. Nothing goes into a shared dataset.