mcp-knowledge-gaps

Find what your knowledge base mentions but doesn't actually explain.

Find concepts mentioned but never defined in your markdown knowledge base (Obsidian vault, Logseq graph, any folder of .md files). Uses fuzzy canonicalization to avoid false positives, ranks gaps by frequency × region-diversity × novelty, generates prioritized research questions, and samples from the long tail via sortition to break confirmation bias in your research queue.

Install

pip install mcp-knowledge-gaps
# or
uvx mcp-knowledge-gaps

Usage

Claude Code

claude mcp add mcp-knowledge-gaps -- mcp-knowledge-gaps

Claude Desktop

Add to claude_desktop_config.json:

{
  "mcpServers": {
    "knowledge_gaps": {
      "command": "uvx",
      "args": ["mcp-knowledge-gaps"]
    }
  }
}

MCP Tools

Tool	Tier	Description
find_gaps	Free	Scan a markdown vault and return concepts mentioned in multiple notes but without their own dedicated note. Applies fuzzy canonicalization and noise filtering.
list_gaps_by_priority	Free	Return gaps ranked by priority: frequency × diversity × novelty (higher = fill this gap first).
generate_research_questions	Pro	Generate prioritized research questions for the top N gaps. Each question comes with a priority score and factor breakdown.
surprise_research_topic	Pro	Sortition sampling — pick a random gap from the LOW-priority long tail. Breaks confirmation bias by surfacing topics you'd never pick yourself.
export_review_queue	Pro	Export a CSV of top-priority gap concepts, suitable for Anki or other spaced-repetition tools. Writes to output_csv and returns the row count.

Pro tier

Unlocks research question generation with RL-weighted ranking, sortition sampling of long-tail gaps, and CSV review queue export.

License activation — any one of these works:

# 1. Environment variable
export KNOWLEDGE_GAPS_LICENSE="eyJhbGc..."

# 2. CLI flag
mcp-knowledge-gaps --license-key "eyJhbGc..."

# 3. Config file
echo "eyJhbGc..." > ~/.mcp-knowledge-gaps/license.jwt

Licenses are verified fully offline — no phone-home, no activation server. Get a license at https://github.com/onetrueclaude-creator/mcp-knowledge-gaps#pro-tier.

Requirements

• Python 3.10+

License

MIT