
Write HTML. Render video. Built for agents.
FrontMCP is a TypeScript framework for building MCP servers with decorators, typed tools, resources, prompts, and agent flows. It includes protocol handling, transport, DI, auth, sessions, and a CLI, so one codebase can run locally and deploy unchanged.
Builders who want to write MCP servers in typed TypeScript and ship them to local or production runtimes.
You can build and deploy a typed MCP server without wiring the protocol, transport, auth, and session machinery by hand.
Use `@FrontMcp`, `@App`, `@Tool`, `@Resource`, `@Prompt`, and `@Agent` with Zod schemas for typed inputs and outputs.
Includes OAuth, JWKS, DCR, per-app auth, and stateful or stateless sessions.
Supports Streamable HTTP and SSE, and can run on Node, Vercel, AWS Lambda, Cloudflare Workers, or a serverless bundle.
Provides `create`, `init`, `dev`, `build`, `inspect`, and `doctor` commands.
Adds cache, remember, approval, codecall, dashboard, feature flags, skilled-openapi, and OpenAPI tool generation.
Includes an E2E testing framework and observability helpers for logging, metrics, and tracing.
# New project (recommended) npx frontmcp create my-app # Existing project npm i -D frontmcp @types/node@^24 npx frontmcp init
The production-grade, TypeScript-first framework for building MCP servers — decorators, DI, auth, and Streamable HTTP, batteries included.
Docs • Quickstart • API Reference • Discord
FrontMCP turns the Model Context Protocol into a
typed, declarative framework. You write clean @Tool, @Resource, and @App
classes; FrontMCP handles the protocol, transport, dependency injection, sessions,
auth, and execution flow — and the same server runs locally and ships to
production unchanged.
import 'reflect-metadata';
import { FrontMcp, LogLevel } from '@frontmcp/sdk';
import HelloApp from './hello.app';
@FrontMcp({
info: { name: 'Demo', version: '0.1.0' },
apps: [HelloApp],
http: { port: 3000 },
logging: { level: LogLevel.Info },
})
export default class Server {}
Node.js 24+ required.
# New project (recommended)
npx frontmcp create my-app
# Existing project
npm i -D frontmcp @types/node@^24
npx frontmcp init
Full setup guide: Installation · Quickstart
Build — decorator-configured @FrontMcp server and @App
domains; typed @Tool, @Resource, and
@Prompt primitives; @Agent multi-step chains; and
scoped Providers / DI.
Secure — Remote & Local OAuth, JWKS, DCR, per-app auth with stateful / stateless sessions (JWT or UUID transport IDs).
Connect & operate — Streamable HTTP + SSE transport,
every MCP protocol revision from 2024-11-05 through
2026-07-28 on one endpoint, capability discovery,
elicitation, hooks, HTTP-discoverable
skills, tool UI / MCP Apps, an in-process
Direct Client (connectOpenAI / connectClaude), and
first-class deployment.
Extend & tooling — official plugins (Cache, Remember, CodeCall,
Dashboard), the OpenAPI adapter, a UI library (HTML/React
widgets, SSR, MCP Bridge), an E2E testing framework, and a
CLI (create, init, dev, build, inspect, doctor).
→ Full reference: docs.agentfront.dev/frontmcp
You install frontmcp (the CLI) and @frontmcp/sdk. Everything else is either
pulled in for you or opt-in.
| Package | Description |
|---|---|
frontmcp | The CLI — create, init, dev, build, inspect, doctor |
@frontmcp/sdk | Core framework — decorators, DI, flows, transport, MCP protocol |
@frontmcp/auth | Authentication, OAuth, JWKS, DCR/CIMD, credential vault |
@frontmcp/testing | E2E test framework with fixtures and matchers |
| Package | Description |
|---|---|
@frontmcp/plugins | Plugin authoring toolkit + official plugin re-exports |
@frontmcp/adapters | OpenAPI adapter — generate tools from an OpenAPI spec |
@frontmcp/skills | Curated SKILL.md catalog for scaffolding and skills install |
@frontmcp/guard | Policy/guard rules for tool inputs and outputs |
@frontmcp/observability | Structured logging, metrics, and tracing helpers |
| Package | Description |
|---|---|
@frontmcp/react | React hooks + client for talking to a FrontMCP server |
@frontmcp/ui | React components, SSR renderers, MCP Bridge |
@frontmcp/uipack | React-free themes, build tools, platform adapters |
| Package | Description |
|---|---|
@frontmcp/edge | Run a server on Cloudflare Workers / V8 isolates from a config |
@frontmcp/storage-sqlite | SQLite-backed session, task, and elicitation stores |
@frontmcp/nx | Nx generators and executors for FrontMCP workspaces |
Published so the packages above resolve, but not intended for direct use:
| Package | Description |
|---|---|
@frontmcp/protocol | The single boundary to the upstream MCP SDK — protocol types |
@frontmcp/di | Dependency injection container |
@frontmcp/utils | Shared utilities — naming, URI, crypto, FS |
@frontmcp/lazy-zod | Lazily-loaded Zod wrapper that keeps cold starts small |
| Package | Description |
|---|---|
@frontmcp/plugin-cache | Cache tool results with a TTL |
@frontmcp/plugin-remember | Per-session memory (this.remember) |
@frontmcp/plugin-approval | Human approval gates before a tool runs |
@frontmcp/plugin-codecall | Let the model compose tool calls as code |
@frontmcp/plugin-dashboard | Built-in web dashboard |
@frontmcp/plugin-feature-flags | Toggle tools and apps at runtime |
@frontmcp/plugin-skilled-openapi | OpenAPI → skills + meta-tools for large APIs |
Keep all @frontmcp/* packages on the same version. A clear "version mismatch" error is thrown at boot if versions drift. (Production Build)
PRs welcome! See CONTRIBUTING.md for workflow, coding standards, and the PR checklist.
Sign in to join the discussion.
No comments yet. Be the first to say what this is good for.

Write HTML. Render video. Built for agents.
Ultra-lightweight, open-source, self-hosted personal AI agent framework in Python with WebUI, tools, memory, MCP, multi-agent workflows, automation, and chat apps
SkillOpt is a text-space optimizer that trains reusable natural-language skills for frozen LLM agents through trajectory-driven edits, validation-gated updates, and deployable best_skill.md artifacts.

Omnigent is an open-source AI agent framework and meta-harness: orchestrate Claude Code, Codex, Cursor, Pi, and custom agents — swap harnesses without rewriting, enforce policies and sandboxing, and collaborate in real time from any device.
A theoretical reconstruction of the Claude Mythos architecture, built from first principles using the available research literature.
🕷️ An adaptive Web Scraping framework that handles everything from a single request to a full-scale crawl!