convez.ai
MCP ServerDados & AnalyticsAvançado
codelined-ag

Extracto

codelined-ag/Extracto

Seu cérebro privado de documentos: PDFs em, RAG em qualquer lugar.

Ver no GitHub
BomPluginScore · de 0 a 100

MCP server que implementa um pipeline completo de ingestão, extração e processamento de documentos PDF com RAG. Auto-hospedável, integrado a múltiplos modelos de visão e armazenamentos vetoriais, com suporte a nuvem (Dropbox, Google Drive, OneDrive) e saída em vários formatos.

Instalação

$npx convez-ai codelined-ag/Extracto

1 comando no terminal: detecta o formato do repositório (marketplace ou plugin avulso) e instala via Claude Code. Requer Node 18+ e o Claude Code instalado.

Métricas

Estrelas
4
Forks
2
Acompanhando
0
Saúde
71%
Último commit
há 2 dias
Versão
v0.11.0
Issues abertas
2
Contribuidores
2

Casos de uso

  • Construir pipelines RAG privados e self-hosted para documentos internos
  • Integrar extração de documentos em agentes Claude via MCP
  • Processar em batch PDFs com OCR, limpeza e armazenamento automático em vector stores

Veredito

Solução robusta para quem precisa de controle total sobre pipelines de documentos sensíveis, com suporte nativo a MCP. A integração com múltiplos LLMs e vector stores oferece flexibilidade real, mas exige expertise em infraestrutura e DevOps.

README

A tradução para o português deste README ainda está na fila.

v1.1.0: cloud integrations end-to-end. Connect Dropbox / Google Drive / OneDrive from the UI (paste your own OAuth client_id+secret if the operator hasn't), browse and import any file from the cloud, send any OCR result back as md, docx, xlsx, obsidian, or zip, and configure watched folders (cloud or local) that auto-submit new files for OCR. See the changelog.

v1.0.0: side-by-side multi-model comparison with server-computed word-level diff, model recommendations from your own OCR history, PII auto-redaction with audit trail, form-field extraction, LaTeX equation extraction, and an E2E encryption scaffold (RSA SPKI public-key registration + AES-256-GCM envelope).


Why

Most document-to-AI tools are SaaS. They cost per page, they see your documents, they lock you into one provider. Extracto is the opposite: one Docker container, your machine, any vision model (local or hosted), output goes wherever you want it. Browser, code, agent, vector store. You pick.


What you get

A complete pipeline from raw document to retrievable knowledge, in one container:

  1. Ingest any PDF, image, watched local folder, or watched Dropbox / Google Drive / OneDrive folder.
  2. Extract with the vision model of your choice (Ollama, Mistral OCR, OpenRouter, any OpenAI-compatible endpoint).
  3. Post-process with a second LLM pass (clean to markdown or strict JSON, with your own instruction).
  4. Chunk + embed + store into Chroma, Qdrant, Weaviate, Milvus, OpenSearch, Pinecone, or Typesense.
  5. Retrieve through a stable v1 REST API, an OpenAI-Chat-Completions adapter, an MCP server, a typed CLI, or the browser UI.
  6. Push any result back to Dropbox / Google Drive / OneDrive, S3/MinIO, or download as md, json, docx, rtf, csv, xlsx, obsidian, or per-page zip.

Everything else (per-user accounts, scoped API keys, rate limits, signed webhooks, S3/MinIO offload, Prometheus metrics, multi-language UI, per-user OAuth credentials when the operator hasn't preconfigured them) is documented at extracto.help.


Quickstart

You need Docker. That's it.

curl -fsSL https://extracto.help/install.sh | bash

Pulls the prebuilt multi-arch image, runs a single container with an auto-generated AUTH_SECRET and a persistent SQLite volume, waits for the healthcheck, and prints the URL. Open http://localhost:3000, sign up, follow the tour.

For the full install (compose stack, Docker + Ollama provisioning, extracto CLI on PATH, Windows path), see extracto.help/install.


Plug everywhere

Same backend, five surfaces. Pick what fits.

SurfaceUse it whenRead
Browser UIYou're a human with a stack of PDFsHow it works
REST API (/api/v1/*)You're building a document-intake pipelineAPI reference
MCP serverYour agent speaks MCP (Claude Desktop, Cursor, Codex, OpenClaw, Hermes)Agents
CLI + SKILL.mdYour agent only has a shell toolSkill file
OpenAI-Chat adapterYou already have OpenAI-SDK code; point it at ExtractoOpenAI compat

OpenAPI 3.1 spec at openapi.yaml. Live Scalar reference at /api/v1/docs on every running instance.


Star history


License

MIT © codelined

Ver o original no GitHub

Bom para

Relacionados