Introducing Connext MCP: Give Your AI Agent Connext Expertise

The way developers write software is changing fast. AI coding agents – tools such as Claude Code, Cursor, and GitHub Copilot – are becoming a core part of the development workflow, generating code, debugging issues, and scaffolding entire projects from natural-language descriptions. But there's a catch: these agents are only as good as the knowledge they have access to. When it comes to specialized middleware such as RTI Connext, even the most capable AI model can stumble, burning time and tokens guessing at APIs it wasn't specifically trained on.

That's why we built Connext MCP. It gives AI coding agents direct, on-demand access to authoritative Connext knowledge – so they can write Data Distribution Service (DDS^®) code that's correct the first time.

What Is Connext MCP?

Connext MCP is an MCP server built on top of the same Connext AI technology that powers Connext Chatbot. MCP – the Model Context Protocol – is an open standard that lets AI agents call external tools during their reasoning process. Think of it as a way for an LLM to pause mid-task, ask a specific question, and use the answer to inform its next step.

The Connext MCP Server provides specialized tools to agents, such as ask_connext_question. When a coding agent hits a question about the Connext API – how to configure a QoS policy, how to use DynamicData, how to set up a WaitSet – it can query Connext AI and get an authoritative answer in real time. That answer feeds directly into the agent's next coding step. No more guessing from training weights. No more wasting tokens on trial and error. This tool and others offered by Connext MCP give agents the specialized knowledge needed to build distributed real-time systems.

Why Does This Matter?

To understand the impact, we ran an experiment. We gave an AI coding agent a real task: build a drone swarm controller that bridges MAVLink telemetry and commands through RTI Connext. We ran the same task twice – once with Connext MCP enabled, and once without.

The results were striking:

• 17% faster completion with Connext MCP enabled.
• Without Connext MCP, the agent spent approximately 66,000 tokens researching the Connext Python API through trial and error – exploring files, reading documentation, and recovering from incorrect API calls.
• With Connext MCP, those same questions were resolved in single, targeted tool calls.

Here's a concrete example of what happened without Connext MCP. The agent tried to access a DDS topic's type through an attribute that doesn't exist in the API. It then launched a lengthy research session – 37 tool calls, 66,000 tokens, and 57 seconds – just to discover the correct method. With Connext MCP, that same question gets a direct, accurate answer in a fraction of the time.

With Connext MCP, your agent will produce code built upon RTI's documentation, code examples, and best practices, ensuring a robust and well-engineered system.

How It Works

Connext MCP plugs into any AI agent that supports the Model Context Protocol. Setup is straightforward: You configure the MCP server in your environment, and your agent gains access to the toolset. From that point forward, whenever the agent encounters a Connext-related question during its work, it can query Connext AI instead of relying on its training data.

The key advantages over a generic AI model working from memory:

• Authoritative Answers: Responses come from RTI's own documentation, API references, code examples, and RTI Academy content – the same knowledge base that powers Connext Chatbot.
• Real-Time Access: The agent queries Connext AI during its reasoning process, not before or after. This means the right information arrives exactly when it's needed.
• No Prompt Engineering Required: The agent decides when and what to ask on its own. You don't need to pre-load context or craft special prompts about Connext.
• Works With Your Existing Tools: Connext MCP integrates with any MCP-compatible agent, fitting into the development workflow you already use.

Get Started

Connext MCP is available today. You can find setup instructions on the Connext AI documentation portal. We'd love to hear how you're using it – reach out to us with your feedback and let us know what you build.

The future of development with Connext is AI-assisted, and we're excited to put that power in your hands.

About the author:

Grayson Honan is the AI Team Lead at RTI, based in the San Francisco Bay Area. Passionate about technology and developer experience, Grayson leverages his extensive technical background to turn emerging AI capabilities into practical tools for RTI’s customers. Grayson holds a BS in Electrical and Computer Engineering from the University of Rochester and an MSE in Embedded Systems from the University of Pennsylvania.