Technical documentation is one of those tasks that rarely fail loudly in projects but often quietly get left behind. A new tool called Blume addresses exactly this point: a single Markdown file is enough to generate complete web documentation from it – including navigation, search function and an MCP connection.

What the tool promises to do

The approach is deliberately lean. The starting point is Markdown, a simple text markup language in which headings, lists and code examples are structured with just a few special characters. From this source file, Blume generates HTML documentation that can be opened in a browser.

Two components turn pure text conversion into a usable portal: navigation that derives from the structure of the document, and a search that lets readers find specific passages. This eliminates the effort that often comes at the start of documentation projects – setting up a page tree, building templates, configuring a search index.

The MCP connection as the real difference

The second output channel is remarkable. Blume also makes the documentation available via MCP. MCP stands for Model Context Protocol, an interface convention through which AI assistants can query external content and tools without having to program a separate integration for each system.

In practice, this means the documentation is not only a website for people but also a queryable source for assistant systems. Anyone working in a development environment or a chat assistant can retrieve documentation content there instead of searching for it in a second window and transferring it via copy and paste.

This dual use is the real point. Until now, documentation portals and AI workflows were separate worlds: on one side a static site generator, on the other content that had to be laboriously transferred into an index or a vector database so that an assistant could work with it. If the documentation itself acts as an MCP source, this intermediate step is eliminated.

Why a single file is an argument

At first glance, maintaining documentation in just one file sounds like a limitation. In many mid-sized projects, it is the opposite. A single text document can be versioned, reviewed in a pull request and edited without a content management system. There are no folder conventions you have to know and no configuration files that nobody understands after six months.

Just as important: the barrier to contributions is lower. Anyone who wants to document an error message or add a deployment step opens a file and writes two paragraphs. Experience shows that this small barrier is precisely what determines whether documentation stays current or becomes outdated after the project is completed.

The limits of the approach are obvious and should be assessed realistically. A single file remains manageable as long as the scope is manageable. For extensive product documentation with multiple language versions, target groups and approval processes, a structured content management system remains the more suitable choice.

Typical areas of use in everyday project work

The cases for which such a lean approach is suitable can be clearly defined:

  • Technical project documentation: architecture decisions, setup instructions and operational notes that live in the repository alongside the code.
  • Internal manuals: processes, responsibilities and checklists for a team that do not need a corporate website but must be searchable.
  • Interface and integration documentation: descriptions that developers need anyway while working in the editor – this is where the MCP connection particularly pays off.
  • Handovers: everything that is transferred to the client or an internal team at the end of a project and is to be maintained there without specialist knowledge.

What needs to be clarified before introduction

Anyone testing such a tool should answer a few organisational questions in advance. Where is the source file located, and who is allowed to change it? Is the HTML output regenerated automatically with every change, or is that a manual step that at some point nobody carries out anymore?

With the MCP connection, the access question is added. Internal documentation often contains details that do not belong in a publicly accessible assistant – server addresses, process details, names of contact persons. It is therefore worth separating from the outset between what is documented publicly and what remains internal.

Assessment

Blume is not a new paradigm but an indication of where documentation tools are heading: away from elaborate portals, towards content that is simultaneously readable for people and queryable for assistant systems. For web and software projects, this means considering MCP as a regular output channel in future – just as HTML and search index are today. Anyone who already maintains documentation as a text file in the repository can add this second channel with little effort and later save themselves the restructuring of grown documentation landscapes.

Sources