MarkIndex
Structure-first search over large Markdown libraries, with no embeddings or vector store
- Pricing
- Open source
- Platforms
- CLI
- Runs
- Locally
- Version
- 0.1.3
- Listing updated
- 4 days ago
Similar MCP servers
About MarkIndex
MarkIndex is an open-source MCP server for knowledge and memory. It runs locally on your machine and is started by your AI client.
Each Markdown file is parsed into a tree of its own sections, then stored with the original lines and a mixed Chinese and English BM25 index in a single SQLite file. An agent walks library, folder, document and section with five read-only tools and reads original text with line numbers, so evidence can be cited as [doc_path, L-].
Indexing is deterministic and does not call a model; LLMs are optional and used only for node summaries, document descriptions and a built-in retrieve() loop through the AI SDK. Keyword search uses Intl.Segmenter segmentation for Chinese and mixed text, with a trigram substring pass. Tool replies are JSON envelopes under a 95,000-character budget, with next-step hints.
It is an independent TypeScript take on the PageIndex idea, unaffiliated with VectifyAI. Requires Node.js 22.12 or newer; index with markindex index ./docs, then serve with markindex mcp ./docs.
Install
Run in a terminal
claude mcp add markindex -- npx -y markindexAdd to claude_desktop_config.json
{
"mcpServers": {
"markindex": {
"command": "npx",
"args": [
"-y",
"markindex"
]
}
}
}Add to ~/.cursor/mcp.json
{
"mcpServers": {
"markindex": {
"command": "npx",
"args": [
"-y",
"markindex"
]
}
}
}Add to .vscode/mcp.json
{
"servers": {
"markindex": {
"type": "stdio",
"command": "npx",
"args": [
"-y",
"markindex"
]
}
}
}Add to ~/.codeium/windsurf/mcp_config.json
{
"mcpServers": {
"markindex": {
"command": "npx",
"args": [
"-y",
"markindex"
]
}
}
}Add to your client's MCP settings (most use this shape)
{
"mcpServers": {
"markindex": {
"command": "npx",
"args": [
"-y",
"markindex"
]
}
}
}Using several servers? Build one config file with the MCP config generator.
What it does
- Section trees
- Parses headings, preface text and headingless documents so every line belongs to one node.
- BM25 plus trigram
- Search handles Chinese, English and identifiers like ISO-27001.
- Citable evidence
- Answers cite document path and line range.
- Incremental indexing
- Rebuilds only changed files, compared by content hash.
- Optional summaries
- Add model-written summaries with rate limits you set.
- Retrieval eval
- Measure retrieval quality on your own questions.
Good for
- Give an agent access to policy or manual libraries
- Search bilingual Chinese and English documents
- Cite exact lines from long filings or guidelines
- Run an agent over a docs folder without a vector database
Pricing
Open source
Free, open source
MarkIndex FAQ
Does it use embeddings?
No. The document's own structure plus BM25 is the index.
Does indexing call an LLM?
No; LLMs are optional for summaries only.
What Node version?
Node.js 22.12 or newer.
Where is the index stored?
In <root>/.markindex/index.db, or a path set with --db; use a local disk.
Is MarkIndex free?
Yes. MarkIndex is open source, so you can use and self-host it for free.
How do I add MarkIndex to Claude Code?
Run this command in your terminal: claude mcp add markindex -- npx -y markindex
What are the best alternatives to MarkIndex?
The closest alternatives to MarkIndex are scree, indexio, jDocmunch MCP, Open Context Engine and Semlith.
Where this comes from
- Description: written from the project's README.
- Install steps: from the official MCP Registry entry.