MCP (Model Context Protocol) servers let you extend Nuveki agents with custom tools that connect to external services. Here's a quick primer.
What Are MCP Servers?
Think of them as plugins. An MCP server is a small service that exposes "tools" — functions that agents can call during a conversation. For example:
- A GitHub MCP server lets agents read repos, create issues, and review PRs
- A Database MCP server lets agents query your database directly
- A Slack MCP server lets agents send messages and read channels
Two Types
Remote (SSE)
- Runs on a remote server
- Connects via HTTP with Server-Sent Events
- Good for: shared team tools, cloud services
Local (STDIO)
- Runs on your machine
- Communicates via standard input/output
- Good for: tools that need local file access, private data
Adding an MCP Server
- Go to Tools in the sidebar
- Click Add Server
- Choose Remote (URL) or Local (command)
- The server's tools are automatically discovered and available to your agents
Security Notes
- Local servers run with your machine's permissions — be mindful of what they can access
- Remote servers go through SSRF protection — private IPs are blocked
- Credentials are encrypted at rest in the vault
Building Your Own
MCP servers can be built in Python or TypeScript. The protocol is straightforward:
- Define your tools (name, description, input schema)
- Implement the handler for each tool
- Expose them via the MCP protocol
Check the tools/ directory in the Nuveki repo for examples of built-in MCP servers.
We're working on a gallery of community MCP servers. If you've built one, share it in Showcase!