my MCP vision for Octomind

In Feb 2022, I changed my LinkedIn to display my vision of the future:

Coding will be a technical discussion between humans & computers. Sooner than we think!

A few months ago, I had to delete the second sentence: Sooner than we think. Because that time is now. Coding will be a technical discussion between humans & computers. Octomind makes sure it is well-tested!

As the only quasi-scalable replicator we have, code will be the key to solving many of humanity’s biggest problems. But only if the code is of high quality and works according to requirements at all times and during all usage scenarios. To guarantee that, we built Octomind.

Friday was a big step in that direction: we listened to your feedback and open-sourced our MCP (Model Context Protocol) implementation so you can start using Octomind from your preferred agent environment: Cursor, Windsurf, VSCode, Claude Desktop and soon other agents of your choice without leaving your preferred interface and workflows. Our goal is to be a good citizen and integrate with your existing workflows. 

Especially since last week’s historic handshake between openAI & anthropic it became clear that MCP is the protocol on which agentic services will cooperate. This cooperation requires safe dissemination of knowledge/context and while MCP still has a few shortcomings regarding authentication & authorization etc., it is on the right trajectory. To address it, major security companies like Cloudflare are allocating >4 of their most senior colleagues to MCP full-time (as of March 2025). Notion, Figma and other scaleups are investing in similar headcounts. We would even go so far and postulate:

‍

Every software company must clearly establish its position regarding MCP.

‍

‍So what’s MCP? It's basically a context or knowledge exchange protocol. Context you have in one app is re-usable and almost 1:1 transferable (might need to be pseudonymized or compressed into a vector) to other MCP-compliant apps. With MCP you are breaking down knowledge silos and it enables you to bridge semantic gaps that the AI beforehands was not able to cross. We saw it in our customers’ AI accuracy: for common-knowledge apps that are consumer-facing like stock trading apps, most AI approaches will have a pretty good understanding. But just within fintech, the domain expertise required can get infinitely complex: the same AI will struggle with high-frequency trading apps that are used by less than 100 expert traders. MCP has the potential to overcome this and give domain expertise to the very same AI that previously struggled with semantic understanding of the app at hand. It can act as a domain expert that teaches & onboards the AI to these difficult to grasp semantic gaps in AI understanding. The best thing about it: this happens in your preferred apps like Cursor, Windsurf and VScode for developers and for product developers: Claude Desktop and soon chatGPT. 

‍

This seamless blend of Cursor and Octomind opens up a whole new dimension of possibilities and use cases. On octomind.dev/mcp, we present 3 of these:

1. Code to test
2. Anything to test
3. Test report and analyses to your preferred app

‍

I think that most of our technical users will use Octomind to generate and execute tests primarily through the Cursor and Windsurf MCP. It’s our preferred way of using it, too :)

Last week’s open-sourcing of our MCP implementation is our contribution to the ecosystem. Watch this video to see it in action.

Let’s together turn testing from afterthought into a first-class citizen. Let’s together repair the damage done by vibe coding!