md-vision MCP server

This is a small helper that lets your AI assistant open Markdown documents stored on your computer, including the images sitting next to the text. It was built mostly for developers building search-over-docs setups, but anyone whose work lives in .md files can find it useful. If your team keeps guides, specs or release notes as Markdown folders, this makes those readable to your AI.

What this MCP server does

When you give your AI a request involving a Markdown document, the AI calls this helper automatically. The helper checks whether the requested file sits within the safe set of folders you pre-approved. Then either lists just the table-of-contents style outline so the AI knows how big the document really is, or pulls out the actual paragraphs together with the illustrations pasted alongside them. Everything lands straight in the reply box, ready for follow-up questions.

Click to zoom

What you can do with it

• Ask your AI to summarise a whole folder of internal wiki-style Markdown pages
• Have diagrams and screenshots shown directly in the answer instead of being skipped
• Get just the 'Deployment' chapter extracted while ignoring hundreds of unrelated lines above and below
• Compare instructions across several version-controlled spec documents side-by-side
• Pull raw GitHub-hosted project READMEs into chat without leaving the editor
• Check quickly whether some old note actually contains the paragraph someone claims it has

Try asking your AI

• “Summarise the architecture overview from ./docs/spec/v2/api-reference.md and include the sequence diagram near the top”
• “What changed between v1 and v2 according to our changelog? Show me both versions' upgrade steps next to each other”
• “Give me the full Troubleshooting section from my personal knowledge-base vault's networking guide”
• “Fetch the contributing guidelines from github.com/acme/widgets/blob/main/CONTRIBUTING.md and tell me whom I should ping for review”

What it gives back to you

For outlines you receive a tidy summary listing each title along with its position in the file, total character count and how many graphics sit underneath it. When asking for real reading material, the response arrives split naturally into prose chunks followed immediately by the corresponding visuals rendered inline. Long reference links remain visible as clickable addresses wherever they were originally written down.

Before you start

Who it's for

Technical writers maintaining large bodies of product documentation, engineers keeping design records beside repository trees, researchers organising literature reviews locally, and anybody else treating collections of annotated Markdown notebooks as primary working assets worth consulting daily via conversational interface.