I built an MCP-native wiki that Claude maintains from your documents
Hi Indie Hackers,
I recently launched LLM Wiki Tools, a web app that turns your documents into a persistent wiki maintained by Claude through MCP.
The problem I wanted to solve is that most AI document workflows feel disposable. You upload a PDF, ask a question, get a decent answer, and then the useful synthesis is stuck inside a chat thread. If you come back later, you often have to ask the model to rediscover the same context again.
I wanted the opposite: every useful interaction should leave something behind.
LLM Wiki Tools has three layers:
- Raw sources: PDFs, office documents, notes, transcripts, web articles, and other files. These stay immutable.
- The wiki: Markdown pages that Claude can create and edit, including overviews, entity pages, concept pages, comparisons, tables, diagrams, citations, and cross-references.
- MCP tools: search, read, write, append, delete, and maintenance operations exposed to Claude, Cursor, custom agents, or scripts.
The workflow is:
1. Create a wiki for a project or topic.
2. Upload source documents.
3. Connect Claude through MCP.
4. Ask Claude to compile, summarize, compare, cite, and maintain the wiki.
Some example prompts:
Read these papers and build a cited overview of the field.
Write entity pages for every model mentioned in these sources.
Lint the wiki and tell me which claims contradict each other.
The product is built for people doing source-heavy work: researchers, students, writers, analysts, engineers, and anyone using AI agents with a lot of reference material.
Pricing is capacity-based rather than token-based. Every plan includes OCR, citations, full-text search, semantic search, and MCP access. Paid tiers mainly increase processed pages, storage, number of wikis, API keys, and upload size.
I am still early and would love feedback from other builders, especially on:
- Whether the positioning makes sense: "file chat" vs "persistent AI-maintained wiki"
- What directories or communities are worth launching in next
- Whether MCP-native workflows are already something people understand, or if I need to explain that more clearly
- Which use cases sound strongest for early users
You can try it here:
Happy to answer questions about the product, the MCP design, the stack, or the launch process.
