Service
AI Integration
Make your product usable by AI assistants.
More and more users get things done in a chat instead of an interface. For your product to take part, it needs a clean API, robust authentication, and capabilities an AI can call reliably. I build that integration and the infrastructure behind it, from the API to the access model.
- API and interface design
- Authentication to current standards
- Integration with AI assistants
- Monitoring and error analysis
Software is becoming a tool for assistants
Users increasingly get things done in a chat instead of an interface. They describe what they want and expect the right application to work in the background. For your product to take part, it needs an interface a model can operate reliably.
That's more than an API. An assistant has to understand which function does what, which data gets changed and which is only read, and what happens when a request fails.
How an integration works
Technically this runs on the Model Context Protocol, or MCP. It describes how an AI assistant discovers and calls external functions. The user states their request, the model picks the right function, your system executes it and returns a result the assistant folds into its answer.
The authorisation in between is what matters. The assistant acts on behalf of one specific user and sees only that user's data. Without that separation an integration wouldn't be defensible.
The highlighted step is the one I build: the interface that decides what an assistant may do on behalf of a user. The dashed line is the return path, where data becomes a phrased answer.
Why this matters now
MCP started as a proposal from Anthropic. It has since become an industry standard: in December 2025 the protocol was donated to the Linux Foundation's Agentic AI Foundation, co-founded by Anthropic, Block, and OpenAI, with support from Google, Microsoft, AWS, Cloudflare, and Bloomberg.
The scale of the ecosystem at the time of that handover: over 10,000 active public MCP servers and more than 97 million monthly downloads of the official developer packages. The protocol is supported by ChatGPT, Gemini, Microsoft Copilot, Cursor, and Visual Studio Code, among others.
In practice that means an integration you build today works not just in one assistant but in all of them that speak the standard. Two years ago that was different, when every platform demanded its own interface.
- 10,000+
- active public MCP servers
- 97M+
- monthly downloads of the official packages
- 75+
- connectors in Claude alone
Figures from the announcement in which Anthropic handed the protocol to the Linux Foundation.
Source: Anthropic, handover to the Agentic AI Foundation, 9 December 2025
What I build
Interface design. Functions cut and documented so a model uses them correctly without guessing, cleanly separated into read and write operations.
Authentication to current standards. OAuth 2.1 with PKCE, so users can connect their account securely without handing over credentials.
Operations and observability. Making error rates, response times, and anomalies visible. With an interface operated by a machine, design flaws otherwise go unnoticed for months.
Frequently asked questions about this service
- Which products is this worth it for?
- For software that performs a recurring task someone would describe to an AI: appointments, querying data, creating or changing something. Where the interface itself is the value, as with visual tools, it brings little.
- What does my product need as a precondition?
- Working data storage and a way to authenticate users securely. Both can already exist or be built alongside. What doesn't work is an interface onto a system that is still unclear internally.
- Isn't this a trend that will pass?
- Possibly. That's why I build the integration on top of a clean API that has value without the AI layer: for your own interface, for partners, for automation. If the standard changes, you swap a layer rather than the foundation.
Other services
Have a similar project in mind?
Tell me about it. I'll reply within 24 hours with an honest assessment.