---
title: "Open Source Tools for Document Agents | Extend"
description: "Open source tools for document agents: the Extend UI Kit, hosted MCP server, CLI, agent skills, ready-made templates, and RealDoc-Bench datasets for real-world documents."
canonical: https://www.extend.ai/resources/open-source
---

# Open Source Tools for Document Agents | Extend

Open source

## Open source tools to make your agents document experts

Extend's OSS helps developers build AI pipelines that can handle the world's hardest docs.

- **Open Tools**: UI Kit, templates, MCP, CLI, and SDKs for your agents. [View tools](https://www.extend.ai/resources/open-source#open-tools)
- **Extend on GitHub**: Source for the UI Kit, CLI, SDKs, skills, and benchmarks. [GitHub](https://github.com/extend-hq)
- **Open Benchmarks**: Public datasets for parsing, extraction, and splitting. [View benchmarks](https://www.extend.ai/resources/open-source#open-benchmarks)

Open Tools

## Everything your agent needs to ship a document pipeline

Start from a working recipe, connect your agent, and ship a review UI for the output. Each tool starts with one command or one link.

### Extend UI Kit

React components for document apps

PDF, DOCX, XLSX, and CSV viewers, bounding box citations, file upload, and e-signature. Add components to your codebase from a shadcn registry and restyle them like your own.

15 components · shadcn registry · MIT

Add the PDF viewer:

```
npx shadcn@latest add https://www.extend.ai/ui/r/styles/new-york/pdf-viewer.json
```

[Browse components](https://www.extend.ai/ui) · [GitHub](https://github.com/extend-hq/ui)

### Extend Templates

Ready-made pipelines for 50+ document types

Each recipe ships a sample document, schema, processor settings, expected output, and a provisioning script. Send one to Claude Code or Codex and adapt the pipeline it creates.

51 recipes · workflow.json · skill.md

Template index for agents:

```
https://www.extend.ai/template-catalog/llms.txt
```

[Browse templates](https://www.extend.ai/templates) · [Bill of lading example](https://www.extend.ai/templates/bill-of-lading)

### Extend MCP

Document tools for any MCP client

One hosted server gives Cursor, Claude Code, Codex, ChatGPT, and any MCP client tools to parse, extract, classify, split, and edit documents. Sign in with OAuth, nothing to install.

86 tools · 9 tool groups · OAuth

Server URL:

```
https://mcp.extend.ai/mcp
```

[Connect your client](https://www.extend.ai/agents/mcp) · [Docs](https://docs.extend.ai/mcp)

### Extend CLI

Document processing from the shell

Parse, extract, classify, split, and edit documents with one command. The setup wizard signs you in and installs a skill, so coding agents can drive the CLI too.

Homebrew, npm, Go · Agent skill included · MIT

Install and sign in:

```
curl -fsSL https://extend.ai/install.sh | sh && extend setup
```

[Install CLI](https://www.extend.ai/agents/cli) · [GitHub](https://github.com/extend-hq/extend-cli)

### Extend SDK

Typed API clients for your stack

Python, TypeScript, Java, and Go clients for parse, extract, classify, split, and edit, with polling and webhook helpers. Add the extend-api skill and your coding agent writes the integration for you.

Python, TypeScript, Java, Go · extend-api skill · MIT

Python:

```
pip install extend-ai
```

TypeScript:

```
npm install extend-ai
```

Go:

```
go get github.com/extend-hq/extend-go-sdk
```

Agent skill:

```
npx skills add extend-hq/extend-agent-plugin --skill extend-api
```

[Set up the SDK](https://www.extend.ai/agents/sdk) · [Docs](https://docs.extend.ai/sdks)

Open Benchmarks

## Public benchmarks with open data

Each benchmark publishes its corpus, metric, and comparison scope, with datasets on Hugging Face so you can reproduce the results or score your own pipeline.

### RealDoc-Bench

Benchmark for document agents

581 real documents from logistics, healthcare, financial services, and real estate, with human-annotated layouts and verified Q&A prompts. Extend Parse 2.0 scores 95.7% field-level Q&A accuracy. Run the harness to score any parser against the same ground truth.

581 documents · 4 industries · CC BY 4.0

Download the documents:

```
hf download Extend-AI/RealDoc-Bench --repo-type dataset
```

[Explore RealDoc-Bench](https://www.extend.ai/resources/realdocbench) · [Dataset](https://huggingface.co/datasets/Extend-AI/RealDoc-Bench) · [GitHub](https://github.com/extend-hq/realdoc-bench)

### LongArray-Extract

Extraction

Tests whether extraction returns complete, schema-faithful arrays when the output grows from a dozen rows to thousands.

**99.2%**: Mean accuracy, Extend

45 financial, clinical, and legal PDFs, up to ~2,200 records

[Dataset](https://huggingface.co/datasets/Extend-AI/LongArray-Extract) · [Read the report](https://www.extend.ai/resources/long-array-extraction-benchmark)

### PoliTax Split

Splitting

Tests boundary detection on long compound tax filings, where frontier models miss subtle boundaries across hundreds of pages.

**+17–44 pts**: Recall lift across models, Extend harness

The 30 largest compound filings in the PoliTax corpus

[Dataset](https://huggingface.co/datasets/Extend-AI/PoliTax-Split) · [Read the report](https://www.extend.ai/resources/document-splitting-benchmark)

[All benchmarks](https://www.extend.ai/benchmarks)