Open-source TypeScript building blocks for AI agents.
A framework for MCP servers • A sandbox for the code agents write • A vector database that runs in memory
FrontMCP docs · FrontMCP · EnclaveVM & VectoriaDB docs · Issues · Releases
| Project | What it is | Docs |
|---|---|---|
| FrontMCP | The TypeScript framework for MCP servers | frontmcp.dev |
| EnclaveVM | A JavaScript sandbox for running the code AI agents write | docs.agentfront.dev |
| VectoriaDB | An in-memory vector database for semantic search, with offline embeddings | docs.agentfront.dev |
Also: mcp-from-openapi, which turns OpenAPI specs into MCP tools, and json-schema-to-zod-v3, which turns JSON Schema into Zod v3 schemas.
FrontMCP is our TypeScript framework for building servers that speak the Model Context Protocol (MCP)—the emerging standard for connecting LLM apps to tools, data, and services. FrontMCP focuses on developer joy and production pragmatism:
- TypeScript-first DX — strong types, Zod schemas, ergonomic decorators
- Composable architecture — plugins, adapters, resources, and tools with clear metadata
- Secure, scoped execution — DI-powered invoker with per-tool access views and consent
- Modern transports — Streamable HTTP, SSE compatibility, and session management
- Great ergonomics — clean logging, hooks, and structured flow control
If you’re building an AI IDE, copilots, or custom workflows, FrontMCP gives you the fastest path from idea → MCP server.
-
The docs are at frontmcp.dev: lessons, the reference, examples and comparisons. Every example runs in your browser.
-
Docs for your agent: FrontMCP's own MCP server gives coding agents those docs, for the FrontMCP release your project uses. See Docs for Your Agent, or add it to Claude Code:
claude mcp add --transport http frontmcp-docs https://frontmcp.dev/mcp
-
Start a server:
npx frontmcp create my-mcp-server