With Latitude Agents, you can design, evaluate, and deploy self-improving AI agents that integrate directly with your tools and data.
We've been working on agents for a while, and continue to be impressed by the things they can do. When we learned about the Model Context Protocol, we knew it was the missing piece to enable truly autonomous agents.
MCP servers were first thought out as an extension for local AI tools (i.e Claude Desktop) so they aren't easily hostable in a shared environment – most only support stdio for comms and they all rely on runtime env vars for configuration.
This meant that to support MCPs for all our users we needed to:
1. Adapt MCPs to support TCP comms 2. Host the MCP server for each of our users
Whenever you create an MCP integration in Latitude, we automatically provision a docker container to run it. The container is exposed in a private VPC only accessible from Latitude's machines.
This gives your MCP out-of-the-box authentication through our API/SDKs.
It's not all wine and roses, of course. Some MCPs require local installation and some manual set up to work properly, which makes them hard for us to host. We are working on potential solutions to this so stay tuned.
We are starting with support for 20+ MCP servers, and we expect to be at 100+ by end of month.
Latitude is free to use and open source, let me know what you think!
