What is the Model Context Protocol (MCP)?
The Model Context Protocol (MCP) is a newly introduced open standard that, in simple terms, provides a common interface for linking AI large language models (LLMs) with external data sources and services in a standardised way.
The problem MCP solves
MCP was developed to solve the complexity of connecting different AI models to different data sources (the “M×N integration problem”). If you wanted your AI assistant to pull information from your company’s documentation, search your codebase, or interact with your project management system, developers had to build and maintain separate integrations for each connection.
MCP eliminates this complexity through a standardised architecture:
- Servers expose data and functionality through a uniform interface
- Clients (AI applications) connect to any compliant server using the same protocol
This means AI assistants can connect with the systems where data lives.
The goal of MCP is ultimately to break down silos and give AI models real-time access to relevant context, so they can produce better, more relevant responses using up-to-date information.
Real-time, context-aware documentation
In the past, we’ve blogged about the promise of personalised content. This is where a Help file or knowledge base provides information that specifically relates to each user’s configuration, abilities, experience, goals and context.
MCP enables documentation that adapts to the user’s current state. A Technical Author can now design documentation systems where:
- AI recognises which version of the software the user is running and provides version-specific guidance
- Documentation automatically reflects the user’s permission level or role
- Troubleshooting steps adjust based on the user’s environment and previous interactions
Connecting AI to Knowledge Bases
MCP allows AI assistants to retrieve specific documentation and data in real time at query time.
In the context of technical documentation, this means:
- Users get answers that are far more accurate and contextually relevant, because the AI can draw on the same sources that a human would consult for the latest information. The AI system doesn’t have to rely on the information stored in its training data.
- The content Technical Authors create can directly power AI responses: the documentation becomes actionable data that the AI will quote or paraphrase when helping users.
- You maintain consistency and a Single Source of Truth between answers from the Knowledge Bases and an AI assistant. There is less risk of information drift. The AI is not hallucinating content on its own; it’s grounded in the single source of truth you maintain.
Implications for technical documentation
Increased documentation value
The value of technical documentation increases because:
- It encourages organisations to keep their documentation comprehensive and up to date. This because both human readers and AI assistants will be using it.
- Improvements in documentation quality immediately translate into better AI responses.
- It amplifies the value of accuracy. If there are gaps or ambiguities in the documents, those will surface in the AI’s output.
Evolving role of Technical Authors
AI systems, in general, want the same as humans: well-structured, clear, and modular information. Plus ça change, plus c’est la même chose. What will change is that someone needs to become an architect of knowledge systems, designing how information flows from sources to users through AI intermediaries. The obvious person to do this is the Technical Author, as they have the communication skills and a user focus.
Using MCP when creating content
Document360 has suggested MCP could also be used to help the Technical Author when they are writing content. MCP could connect an AI assistant to internal document repositories and create draft content whenever there are changes to code or source documents. Of course, the writer would need to review the changes, but the heavy lifting of finding what changed and where to put it can be handled by the AI.
Semantic enrichment
MCP’s architecture emphasises structured data and clear interfaces. Content may need richer metadata to help AI systems understand context, relationships between topics, and the appropriate use cases for different information.
Metadata is information about information. For example, it can help the AI system know which product (e.g. Cloud or on-premise) and version (e.g. version 1.1 or version 1.2) the information is describing.
The best approach for doing this is currently up for debate:
- HTML and XML offer semantic richness but consume more AI context memory, potentially increasing hallucinations
- Markdown provides compactness but lacks robust metadata capabilities
- AsciiDoc might gain popularity as a middle ground
Anticipating what the AI system will ask
Technical Authors might need to anticipate common questions and structure content to align with how MCP servers expose data through prompts and resources.
Connecting more content together
MCP makes it feasible to create AI systems that synthesise information from multiple sources. These might be unexpected sources. This could enable Technical Authors to:
- Identify documentation gaps by analysing what questions require information from multiple sources
- Create unified experiences that combine official documentation with community knowledge
Managing the risks
Data security concerns
MCP functions like a master key that lets AI access many different information sources. Therein lies risks. If someone gets control of an AI that has MCP connections, they suddenly have access to everything that AI can reach. If you don’t carefully control what each MCP connection can access, you might accidentally give the AI (and anyone using it) more information than they should have.
Third-party MCP servers
Anyone can create an MCP server. But how do you know if that server is secure and trustworthy?
If you connect your AI to a poorly secured or malicious MCP server:
- That server might steal the queries you send (revealing what you’re working on)
- It could send back malicious content (prompt injection attacks)
- It might log and keep sensitive information you share
- Your credentials could be compromised
Quality Assurance challenges
When users interact with documentation through AI rather than reading it directly, the organisation faces new quality challenges:
- It needs to test not just whether their documentation is clear when read directly, but also whether AI systems interpret and synthesise it correctly when responding to queries.
- If MCP servers are pulling from multiple documentation sources (such as product documentation, API specifications, and release notes), the writers must ensure consistency across all these sources, as AI will potentially combine them.
- In traditional documentation, some ambiguity might be acceptable when humans can use judgement. With AI mediation, ambiguous phrasing could lead to inconsistent or incorrect responses.
Attribution and transparency
When users receive AI-generated responses that have been synthesised from the documentation, they might not know what source material informed the answer. Technical Authors should consider:
- How to ensure proper attribution and version transparency (this relates back to semantic content)
- When AI synthesis is appropriate, versus when users need to read original documentation
- How to maintain user trust when they can’t easily verify the AI’s sources
Editorial control
How do you maintain style consistency, brand voice, and messaging control when content is synthesised on-the-fly? There will be a need for guardrails, templates, and other editorial quality criteria to address this.
Looking forward
The Model Context Protocol is still new technology. It illustrates where technical documentation is heading. Organisations will be designing documentation not just for human readers, but for the AI systems that will increasingly mediate between documentation and users.
The goal remains the same: helping users accomplish their tasks. The methods, however, are evolving rapidly, and Technical Authors have an opportunity to shape how this evolution unfolds.
If you need training or guidance on using AI in technical documentation, Cherryleaf can help.
Leave a Reply