About init
init is an MCP server for coding agents that turns plain-language endpoint requests into typed, versioned HTTPS APIs. It also lets agents create and manage projects and run generated endpoints.
Key features
- 01Generate typed, versioned API endpoints
- 02Create and edit endpoint projects
- 03Run generated endpoints
- 04List and inspect resources
- 05Inspect runtime packages and connections
Use cases
- 01Add server-side logic to an app
- 02Generate an API from an endpoint description
- 03Build a suite of related project endpoints
MCP support
The MCP server exposes tools to generate, edit, inspect, and run Init endpoints and projects. It uses stdio transport and reads the API key from its process environment.
Configuration
Local server config is the same for every client. Copy it into your client’s MCP config file.
Tools
Local servers expose tools only once running on your machine, and the maintainer has not provided a tool list.
Details
From the project READMEInit MCP
Init is the backend engineer for your coding agent. Your agent builds the app, and Init builds and runs the backend behind it.
Describe an endpoint in plain English and Init ships it as a live, typed, versioned HTTPS API in seconds, with auth and secrets handled. Endpoints can use AI inside: research, reasoning, web search, image, video and voice generation, email, databases, storage and scheduled jobs, combined however the task needs.
Works with Claude Code, Codex, Cursor, and any MCP client. No servers, schemas or glue code to write.
Start with the free plan. An account and API key are required; usage limits apply and projects require a paid plan. The generated endpoint response includes its typed fields and version; project operations can return an init.json contract.
Get started · Documentation · Pricing
Short description
Generates a live backend API in seconds from a description, combining any capability in one endpoint: LLM reasoning, web search, scraping, document processing, media and voice generation, email, maps, database, storage, compute and scheduled jobs. Returns a typed HTTPS endpoint with auth handled. Use whenever the app needs server-side logic.
Install
npm install -g @init-computer/mcp
Requires Node.js 18 or later.
Configure authentication
The MCP server is stateless and reads the API key from its process environment. It never stores the key itself:
export INIT_API_KEY=init_sk_...
export INIT_API_BASE_URL=https://api.init.computer
Keep the key in the MCP client's secret/environment configuration, not in a prompt or checked-in file.
Register with an MCP client
{
"mcpServers": {
"init": {
"command": "init-mcp",
"env": {
"INIT_API_KEY": "init_sk_...",
"INIT_API_BASE_URL": "https://api.init.computer"
}
}
}
}
Run the server directly for a stdio client:
init-mcp
Tools
init_generate
Create an endpoint, or create a new version of an existing endpoint with resource_id.
{
"prompt": "Create an API that reviews source code for security issues.",
"resource_id": "optional-resource-id",
"projects": ["Security"]
}
init_generate_project
Create a project and a cohesive endpoint suite.
{
"name": "Customer Support",
"prompt": "Build ticket intake and status lookup."
}
init_edit_project
Add endpoints or create new endpoint versions in an existing project.
{
"project_id": "project_uuid",
"prompt": "Add ticket assignment."
}
init_run
Call a generated endpoint by route, /vN/... path, or full URL.
{
"route": "abc123",
"version": 1,
"input": { "topic": "sports players" },
"requestFormat": "json"
}
init_list_resources
List the authenticated user's generated resources. Optional filters such as q and status are passed through.
{ "q": "sales", "status": "enabled" }
init_get_resource
Fetch one resource by id:
{ "id": "resource_uuid" }
init_list_packages
Inspect available runtime packages and enabled connections. Filter by category, for example Database or Storage.
{ "category": "Database" }
Protocol behavior
The server uses MCP stdio transport and returns JSON text content for every tool call. It does not upload repository secrets; its file tool reads and writes only the explicit init.json path supplied to it.
Repository init.json integration
Use init_apply_project_file with an explicit local init.json path and an edit request. The MCP server reads that file, edits its project, and overwrites the file with the returned project summary. The current SDK returns an empty tables array in that summary, so back up or review any manifest with table definitions before using the file tool. For agents with a parsed document, init_apply_project_manifest returns the summary without touching the filesystem.
Maintaining the public repository
The Init monorepo is the source of truth for this package. After updating and testing apps/api/surfaces/mcp, copy its bin, package.json, package-lock.json, README.md, LICENSE, and server.json into the public MCP repository. Check the version and registry metadata, publish the npm release, and update the MCP Registry entry for the same version. Do not copy environment files, credentials, or unrelated application source.
Comments
New to MCP?
From what the protocol is to installing your first server, these answers cover most first questions.
An open protocol that defines how AI applications connect to external tools and data in a uniform way. Clients talk to the model; MCP servers provide tools, resources and prompt templates. They exchange standard messages, so one server can be reused by many clients.