Mind Forge
This server does not currently have a deployable release on Glama. Follow the repository's setup instructions to install and configure it for your MCP client, including any required credentials and local dependencies.
Mind Forge
Ingest, query, and generate study materials from documents — all through your OpenCode chat.
Mind Forge is an OpenCode plugin that turns documents (PDFs, DOCX files, Markdown, images, web pages) into a searchable knowledge base with vector search, a knowledge graph, and study tools. You describe what you want in chat, and the LLM calls the right MCP tool automatically.
Status: MVP implemented. The full pipe — ingest → embed → graph → study — is functional.
Quick Start
# 1. Clone and install
git clone https://github.com/goncalompontes/mind-forge.git
cd mind-forge
npm install
npm run build
# 2. Register in your OpenCode configAdd to your opencode.json:
{
"mcpServers": {
"mind-forge": {
"command": "node",
"args": ["/path/to/mind-forge/dist/index.js"]
}
}
}Then use it in chat:
You: Ingest the PDF at ~/papers/transformer-attention.pdf
Mind Forge: Ingested "Attention Is All You Need" (PDF, 15 chunks, 42 entities, 18 relationships)
You: Query: "how does multi-head attention work?"
Mind Forge: [3 results, scores 84–92%] Found in "Attention Is All You Need" chunk 4: "Multi-head attention allows the model to jointly attend to information from different representation subspaces..."
Related MCP server: Factograph MCP
Architecture
Mind Forge registers three MCP tools that the LLM calls automatically:
Tool | Purpose | Pipeline |
| Import a document | extract → embed → store → graph |
| Search your knowledge base | vector search + graph enrichment + FTS5 |
| Create study materials | cards, quiz, exam, or review |
Data Flow
Document → extract() → chunks → embed() → store (SQLite + sqlite-vec)
↘ extractEntitiesAndRelationships() → graph store
↓
User query → embed() → vector search (ANN) → merge with FTS5 + graph enrichment → results
↓
User request → createCards() / generateQuiz() / createExam() → study materialsStorage
SQLite via
better-sqlite3with WAL modeVector index via
sqlite-vec(768-dimension FLOAT embeddings)Full-text search via FTS5 virtual table
Knowledge graph in SQLite (entities + relationships tables)
Single file at
~/.mind-forge/store.db(configurable viaMIND_FORGE_DB_PATH)
Source Format Support
Format | Extractor | Library | Notes |
|
| Metadata via | |
DOCX |
|
| Metadata from |
Markdown |
|
| Frontmatter parsing (title, author, custom fields) |
Image |
|
| PNG, JPG, JPEG, WebP; configurable OCR language |
URL |
|
| SSRF protection, size-limited streaming |
Configuration
Mind Forge auto-detects the best embedding provider. You can configure via environment variables:
Env Variable | Purpose | Default |
| Database file path |
|
| Ollama server URL |
|
Embedding provider selection (via EmbeddingConfig):
auto(default) — tries Ollama first, falls back to API provider if configuredollama— local Ollama (nomic-embed-textdefault, falls back toall-minilm,mxbai-embed-large)llm— OpenAI-compatible API (requiresapiKey)
Default chunk size: 1000 tokens (~4000 characters), paragraph-aware splitting.
Project Structure
src/
├── index.ts # Plugin entry point — registers MCP server
├── types.ts # All shared domain types (12 interfaces, 5 type aliases)
├── embed/ # Embedding providers
│ ├── provider.ts # Factory — auto/Ollama/LLM selection
│ ├── ollama.ts # Ollama adapter (ollama npm package)
│ └── llm-provider.ts # OpenAI-compatible API adapter
├── extract/ # Document extraction
│ ├── index.ts # Orchestrator + paragraph-aware chunking
│ ├── pdf.ts # PDF via pdftotext + pdf-parse
│ ├── docx.ts # DOCX via mammoth
│ ├── markdown.ts # Markdown via gray-matter
│ ├── image.ts # Image OCR via tesseract.js
│ └── url.ts # Web pages via @mozilla/readability
├── store/ # SQLite persistence
│ ├── database.ts # Singleton, schema, sqlite-vec init
│ ├── documents.ts # Document + chunk CRUD
│ └── vectors.ts # Vector insert + ANN search
├── graph/ # Knowledge graph
│ ├── extractor.ts # Pattern-based entity/relationship extraction
│ ├── index.ts # Graph storage (atomic transactions)
│ └── query.ts # BFS traversal, neighbors, pathfinding
├── study/ # Study tools
│ ├── cards.ts # SM-2 spaced repetition cards
│ ├── quiz.ts # Quiz generation + grading (MCQ, T/F, fill-blank)
│ └── exam.ts # Timed exam mode
└── mcp/ # MCP server
├── server.ts # Server registration + 3 tool handlers
├── ingest.tool.ts # IngestTool class (extract → embed → store → graph)
└── query.tool.ts # QueryTool class (hybrid search)Dependencies
Package | Purpose |
| MCP server framework |
| OpenCode plugin registration |
| SQLite database |
| Vector search extension |
| Local embedding via Ollama |
| Image OCR |
| Web page content extraction |
| DOCX text extraction |
| Markdown frontmatter parsing |
| PDF text extraction (fallback) |
| DOM parsing for Readability |
Scripts
Script | Command |
|
|
|
|
|
|
Key Design Decisions
Conversational interface: All interaction through OpenCode chat via MCP tools. No slash commands, no custom UI.
Graceful degradation: Embedding or graph failures don't block ingestion. Document + chunks are always stored.
Hybrid search: Vector similarity (0.7 weight) + FTS5 BM25 (0.3 weight) merged with dedup by chunk ID.
Pattern-based extraction: Regex patterns for entities and relationships at MVP (LLM callback extension point available).
SSRF protection: URL extraction resolves hostnames to IPs and rejects private/reserved ranges before connecting.
Tool Schema Changelog
Recent tool additions, removals, and schema changes observed during successful MCP inspections.
Loading tool schema changelog…
This server cannot be deployed
Maintenance
Related MCP Connectors
Convert PDFs, Office docs, websites and audio into AI-ready Markdown, then search and ask your docs.
Turn any recording or document into notes, flashcards and quizzes your agent can read and act on
Build study flashcards and exam-prep decks from your AI chat, all stored locally.
Ingest, manage, and retrieve documents for RAG-powered AI applications
Related MCP Servers
- FlicenseNot gradedqualityNot gradedmaintenanceEnables personalized AI tutoring by allowing students to upload PDF/DOCX study materials that are processed and indexed for semantic search. Provides intelligent responses based on the student's own learning materials using RAG technology.-
- FlicenseNot gradedqualityDmaintenanceEnables building and exploring a knowledge graph from files (PDF, DOCX, code) with AI enrichment via Ollama or Claude. Supports pinning documents, linking entities, and traversing the graph.-
- AlicenseAqualityDmaintenanceEnables browsing and preparing study materials from local documents (PDF, Markdown, text, Word) via tools for overview, search, reading, chunking, and quiz generation.9MIT
- AlicenseNot gradedqualityDmaintenanceConverts study materials like PDFs, slides, and DOCX into a searchable vector database with semantic search, summarization, and flashcard generation, integrated with Notion.5MIT