Open Source MCP Apps
Browse open source Model Context Protocol (MCP) servers, gateways and platforms that connect AI agents like Claude and Cursor to your tools, data and apps.
Apps in MCP
Fortress
Automation · Browser · Developer Tools
A stealth Chromium engine that corrects the browser fingerprint inside the C++ engine itself, so scrapers and browser agents stop getting blocked, with a 29-tool MCP server for AI agents.
Stealth Browser MCP
Developer Tools · Mcp
An MCP server giving AI agents real, undetectable browser automation — built on nodriver and Chrome DevTools Protocol to navigate sites protected by Cloudflare and other anti-bot systems for automation and testing workflows.
About MCP
What MCP is
The Model Context Protocol (MCP) is an open standard for connecting AI models to the outside world. An MCP server exposes tools, resources and prompts over a common interface, and any MCP client — Claude, Cursor, Codex, Gemini CLI, LibreChat and many more — can call them without a custom integration for each pairing.
This category collects open source projects where MCP is central to what they do: standalone MCP servers that give agents a specific capability, gateways and platforms that host, secure and govern many servers at once, and tools that turn existing software into something an agent can drive.
What you’ll find here
- Capability servers — a single, focused job: semantic code search, browser automation, a knowledge graph of your codebase, a news feed.
- Gateways and proxies — sit between agents and servers to handle authentication, access control, rate limits and observability across a fleet of MCP endpoints.
- Platforms and registries — self-hosted places for an organization to publish, discover and monitor the MCP servers its teams use.
- MCP-native apps — products built around exposing their data and actions to agents first.
How to choose
Transport and deployment. Local servers usually run over stdio next to the client; remote servers speak Streamable HTTP and need hosting, TLS and auth. Check which the project supports and whether that fits where your agents run.
Authentication and scope. An MCP server acts with whatever permissions you give it. Look for OAuth support, per-tool scoping and read-only modes before pointing one at production data.
Tool surface. More tools isn’t always better — every tool description costs context in the model’s window. Prefer servers with a tight, well-described set of tools for the job.
Client compatibility. Most servers work with any spec-compliant client, but some features (resources, prompts, sampling, elicitation) are unevenly supported. Confirm your client handles what the server relies on.
Maintenance. The protocol is still moving. A server that tracks recent spec revisions and has active releases will age far better than a weekend prototype — the repository health score on each listing is a quick way to check.
