Key takeaway
As of April 15, 2026, the practical question is no longer 'which PDF app is best' but 'which PDF skills can be installed, audited, and productionized inside real agent stacks.'
Core judgment
- As of April 15, 2026, OpenClaw and Trae Agent clearly ship open-source code; Claude Code now has an official GitHub repo; Codex CLI is open-source, while the Codex app and cloud remain managed OpenAI product surfaces.
- For PDF workflows, CLI / Python / Java libraries are the most portable install form across all four agent families. MCP is best-documented in Claude Code and Trae, while Codex and OpenClaw benefit from treating MCP as a wrapper layer rather than the only dependency.
- Desktop GUI PDF tools are not unusable for agents, but they depend more on browser or desktop automation and are typically weaker on stability and auditability than CLI or API routes.
Agent Install
How OpenClaw, Claude Code, Codex, and Trae install PDF skills
Everything below is tied to what was publicly verifiable on April 15, 2026, and distinguishes native support from wrapper-based support and UI-first workarounds.
Codex
Install the CLI or IDE first, then package PDF tooling with AGENTS.md and skills
Best for: PDF automation inside codebases, batch scripts, document understanding, and translation-delivery workflows
Installation pattern
- Install Codex CLI with `npm install -g @openai/codex` or `brew install --cask codex`.
- Add `AGENTS.md` at the repo root and document the PDF workflow, test commands, and permission boundaries.
- Package PDF tooling as repo scripts such as `tools/pdf/run_ocr.sh` or `tools/pdf/parse_docling.py`.
- If you use Codex skills in the app, CLI, or IDE, bundle instructions, resources, and scripts as reusable skills.
Capabilities and limits
- Public documentation now confirms both Skills and AGENTS.md.
- The CLI is open-source, while the app and cloud remain managed product surfaces.
- The safest integration path for PDF skills is still repo-level scripts plus AGENTS.md.
Claude Code
Four official extension paths: skills, plugins, MCP, and CLAUDE.md
Best for: Teams that want the clearest official extension model and reusable PDF skills as durable team assets
Installation pattern
- Install Claude Code with `curl -fsSL https://claude.ai/install.sh | bash`, or use Homebrew / WinGet.
- Native skills live in `~/.claude/skills/<skill>/SKILL.md` or project-level `.claude/skills/<skill>/SKILL.md`.
- Use `claude mcp add ...` to connect local stdio servers, remote HTTP servers, or OAuth-backed tools.
- Bundle skills, agents, hooks, and MCP servers into plugins when you need a sharable team distribution format.
Capabilities and limits
- Its public documentation is the most explicit across skills, plugins, and MCP.
- It works well for both native skills and MCP wrappers around OCR, parsing, translation, and RAG services.
- The core CLI has an official GitHub repo, while the model and service layer remain proprietary.
OpenClaw
Workspace skills, plugins, ClawHub, and a gateway make it the closest thing to a personal agent OS
Best for: Personal AI assistants, multi-channel automation, and mixed browser / desktop / shell PDF workflows
Installation pattern
- Install OpenClaw from the official repo or install script; the main runtime entry is the Gateway.
- Workspace skills live in `~/.openclaw/workspace/skills/<skill>/SKILL.md`.
- Use plugins when a PDF workflow also needs channels, host integrations, or system capabilities.
- If ClawHub is enabled, agents can search and fetch skills, but production setups should still whitelist and review them.
Capabilities and limits
- The official README documents workspace roots, skills paths, and ClawHub behavior clearly.
- It is stronger than pure IDE agents when browser, desktop, or host-command automation matters.
- It is also the most open, which means permission control and supply-chain review matter more.
Trae
Trae IDE uses Agent Skills, @Agent, and MCP; the OSS Trae Agent uses YAML plus MCP
Best for: IDE collaboration, multi-agent orchestration, and research-oriented or extensible coding-agent setups
Installation pattern
- Install Trae IDE or SOLO from the official download page when you want the desktop product surface.
- The official Trae blog already documents Agent Skills creation, import, usage, and MCP support through `@Agent`.
- For the open-source agent path, use `git clone https://github.com/bytedance/trae-agent.git && uv sync --all-extras`.
- Add `mcp_servers` in the `trae-agent` config to attach external PDF skills and document tools.
Capabilities and limits
- Trae Agent has an official MIT-licensed GitHub repo.
- The Trae IDE and SOLO product surfaces publicly point to Agent Skills and MCP usage.
- Use the open agent when you need tighter control; add the IDE when visual workflows matter.
Compatibility
Which agents can really use which PDF skill packaging patterns
Do not confuse 'supports PDFs' with 'supports every PDF skill.' Real compatibility depends on packaging form: native skills, repo rules, CLI/libraries, MCP, SaaS APIs, or desktop GUI automation.
| Install form | Codex | Claude Code | OpenClaw | Trae | Verdict |
|---|---|---|---|---|---|
| Native skills / commands | Direct | Native | Native | Direct | Claude Code is the clearest today; OpenClaw has workspace skills; Trae has Agent Skills; Codex publicly confirms Skills but exposes fewer file-system details. |
| Repo rules files (AGENTS.md / CLAUDE.md / Rules) | Native | Direct | Direct | Direct | All four agent families can consume this layer; it is the most portable and least coupled way to inject team knowledge. |
| CLI / Python / Java libraries | Direct | Direct | Direct | Direct | This is the most reusable packaging form across agent families and the best first layer to deploy. |
| MCP server | Wrapper | Native | Wrapper | Direct | Claude Code is strongest natively; Trae also points clearly to MCP; Codex and OpenClaw usually benefit from MCP through wrappers, plugins, or gateways. |
| SaaS API / cloud service | Direct | Direct | Direct | Direct | All four agent families can use this layer reliably when API keys are governed and packaged as tools or scripts. |
| Desktop GUI / RPA | Limited | Limited | Direct | Wrapper | OpenClaw is friendlier to browser and desktop control; Codex and Claude Code should not treat GUI automation as the primary path. |
Catalog
36 PDF skills / tools with open-vs-closed status, GitHub, and installability
Here, 'skills' are concrete installable building blocks: open-source libraries, CLIs, MCP servers, SaaS APIs, and desktop products. Open-source rows include GitHub.
| Skill / Tool | Category | Open vs closed | Install form | GitHub / official | Best for | Note |
|---|---|---|---|---|---|---|
| Tesseract OCR | OCR | Open | CLI / library | General OCRMultilingual OCR | Strong local open-source baseDepends on preprocessing quality | |
| OCRmyPDF | OCR | Open | CLI / library | Searchable PDF outputAgent preprocessing | Strong local open-source baseCommon in production pipelines | |
| PaddleOCR | OCR | Open | CLI / library | Multilingual OCREnterprise forms and contracts | Strong in Chinese-heavy workflowsCommon in production pipelines | |
| docTR | OCR | Open | CLI / library | General OCREnterprise forms and contracts | Research-friendlyDepends on preprocessing quality | |
| Docling | PDF parsing | Open | CLI / library | LLM-ready structuringComplex layouts | Useful as pipeline infrastructureWorks especially well with MCP | |
| docling-mcp | PDF parsing | Open | MCP | MCP integrationLLM-ready structuring | Works especially well with MCPUseful as pipeline infrastructure | |
| GROBID | PDF parsing | Open | CLI / library | Academic papersResearch and technical PDFs | Research-friendlyCommon in production pipelines | |
| Nougat | PDF parsing | Open | CLI / library | Academic papersFormula-heavy documents | Research-friendlyNot a general-purpose OCR tool | |
| MinerU | PDF parsing | Open | CLI / library | Complex layoutsFormula-heavy documents | Strong on complex layoutsCommon in production pipelines | |
| PyMuPDF | PDF operations | Open | CLI / library | High-performance runtimeLightweight PDF operations | Common in production pipelinesUseful as pipeline infrastructure | |
| PyMuPDF4LLM | PDF operations | Open | CLI / library | Agent preprocessingLLM-ready structuring | Useful as pipeline infrastructureCommon in production pipelines | |
| pypdf | PDF operations | Open | CLI / library | Lightweight PDF operationsPDF structure operations | Pure Python friendlyUseful as pipeline infrastructure | |
| pdfplumber | Table extraction | Open | CLI / library | Table debuggingText-based tables | Good for debuggingUseful as pipeline infrastructure | |
| Unstructured | Document ETL | Open | CLI / library | Document chunkingAgent preprocessing | Useful as pipeline infrastructureGood for team workflows | |
| unstructured-api | Document ETL | Open | SaaS API | Internal API layerDocument chunking | API-firstGood for team workflows | |
| Tabula | Table extraction | Open | CLI / library | Text-based tablesBatch table extraction | Weak on noisy scansCommon in production pipelines | |
| tabula-java | Table extraction | Open | CLI / library | Batch table extractionJava enterprise stacks | Common in production pipelinesUseful as pipeline infrastructure | |
| qpdf | PDF operations | Open | CLI / library | PDF structure operationsBatch post-processing | Common in production pipelinesUseful as pipeline infrastructure | |
| pdfcpu | PDF operations | Open | CLI / library | Batch post-processingPDF structure operations | Common in production pipelinesUseful as pipeline infrastructure | |
| Apache PDFBox | PDF operations | Open | CLI / library | Java enterprise stacksPDF structure operations | Common in production pipelinesGood for team workflows | |
| OpenAI PDF Files | RAG / reasoning | Closed | SaaS API | PDF reasoningCross-document search | API-firstBetter for reasoning than layout fidelity | |
| OpenAI File Search | RAG / reasoning | Closed | SaaS API | Cross-document searchTeam knowledge search | API-firstGood for team workflows | |
| Claude PDF Support | RAG / reasoning | Closed | SaaS API | PDF reasoningResearch and technical PDFs | API-firstBetter for reasoning than layout fidelity | |
| Claude Citations | Knowledge Q&A | Closed | SaaS API | Grounded answersTeam knowledge search | API-firstGood for team workflows | |
| Mistral OCR | Enterprise document AI | Closed | SaaS API | Cloud OCR APIComplex layouts | API-firstAdds vendor cost and dependency | |
| Mathpix PDF to Markdown | PDF parsing | Closed | SaaS API | Formula-heavy documentsAcademic papers | Research-friendlyAdds vendor cost and dependency | |
| Google Document AI | Enterprise document AI | Closed | SaaS API | Enterprise forms and contractsInternal API layer | Enterprise-orientedAPI-first | |
| Azure Document Intelligence | Enterprise document AI | Closed | SaaS API | Enterprise forms and contractsCloud OCR API | Enterprise-orientedAPI-first | |
| Amazon Textract | Enterprise document AI | Closed | SaaS API | Enterprise forms and contractsCloud OCR API | Enterprise-orientedAPI-first | |
| Adobe Acrobat AI Assistant | Desktop PDF | Closed | Desktop GUI / RPA | Desktop reviewTeam knowledge search | GUI-firstOften needs wrapper automation | |
| Adobe Translate PDF | Translation | Closed | Desktop GUI / RPA | Desktop translation workflowsMultilingual delivery | GUI-firstHigh-value translation layer | |
| ABBYY FineReader PDF | Desktop PDF | Closed | Desktop GUI / RPA | Desktop OCR and reviewSearchable PDF output | GUI-firstEnterprise-oriented | |
| Nanonets | Invoice automation | Closed | SaaS API | Invoices and receiptsInternal API layer | API-firstEnterprise-oriented | |
| Rossum | Invoice automation | Closed | SaaS API | Invoices and receiptsEnterprise forms and contracts | Enterprise-orientedAPI-first | |
| Parseur | Template extraction | Closed | SaaS API | Template-driven extractionInternal API layer | API-firstCommon in production pipelines | |
| Reflo | Translation | Closed | SaaS API | Multilingual deliveryDesktop translation workflows | High-value translation layerStrong on complex layouts | |
| DeepL Files + Glossary | Translation | Closed | SaaS API | Termbase-driven translationMultilingual delivery | High-value translation layerGood for team workflows | |
| Smallpdf Translate PDF | Translation | Closed | Desktop GUI / RPA | Quick consumer translationDesktop translation workflows | GUI-firstOften needs wrapper automation | |
| iLovePDF Translate PDF | Translation | Closed | Desktop GUI / RPA | Quick consumer translationDesktop translation workflows | GUI-firstOften needs wrapper automation | |
| PDFgear ChatPDF | Knowledge Q&A | Closed | Desktop GUI / RPA | Desktop chat with PDFPDF reasoning | GUI-firstOften needs wrapper automation | |
| UPDF Chat with PDF | Knowledge Q&A | Closed | Desktop GUI / RPA | Desktop chat with PDFPDF reasoning | GUI-firstOften needs wrapper automation | |
| AskYourPDF | Knowledge Q&A | Closed | SaaS API | PDF reasoningTeam knowledge search | API-firstBetter for reasoning than layout fidelity | |
| Humata | Knowledge Q&A | Closed | SaaS API | Team knowledge searchCross-document search | API-firstGood for team workflows |
Solution
A production solution is a stack, not a shopping list
A deployable PDF-agent solution combines an agent, PDF skills, a packaging layer, permission controls, and sample-document regression tests. Buying isolated tools is not enough.
Blueprint A: Local-first open-source PDF agent baseline
Best fit: Privacy-sensitive teams that want control, lower vendor dependency, and are willing to operate their own stack
Recommended stack
- Agent: Claude Code or OpenClaw, with Trae Agent OSS as a strong alternative
- OCR: Tesseract + OCRmyPDF + PaddleOCR
- Parsing: Docling / MinerU / GROBID / Nougat
- Operations: PyMuPDF + pypdf + qpdf + pdfcpu
- Tables: pdfplumber + Tabula / tabula-java
Implementation steps
- Install PDF capabilities first as CLI tools and Python scripts instead of starting with GUI products.
- Package those scripts as reusable skills for each agent family: `.claude/skills`, OpenClaw workspace skills, Trae Agent Skills or YAML, and Codex repo scripts plus AGENTS.md.
- Prepare 5 to 10 sample documents per document type and run regression checks for OCR, tables, formulas, and reading order.
Main risks
- Self-hosted stacks cost more to maintain than SaaS layers.
- Accuracy can drop on complex layouts and low-resource languages.
- Permissions, logging, and regression governance remain your responsibility.
Blueprint B: Enterprise API-centered PDF agent platform
Best fit: Enterprises that already run cloud infrastructure and care about SLA, auditability, identity, and compliance
Recommended stack
- Agent: Claude Code or Trae, with Codex covering the code and automation layer
- OCR / extraction: Google Document AI / Azure Document Intelligence / Amazon Textract
- Knowledge layer: OpenAI PDF Files + File Search or Claude PDF + citations
- Business flow tools: Nanonets / Rossum / Parseur
- Post-processing: qpdf / pypdf / PyMuPDF
Implementation steps
- Wrap closed cloud services behind internal APIs or MCP wrappers instead of wiring every vendor directly into the agent.
- Route contracts, invoices, research PDFs, and branded collateral through different queues rather than sharing a single prompt chain.
- Put permissions and audit controls in the orchestration layer, not inside prompts.
Main risks
- Vendor lock-in and cost growth remain real risks.
- API output structures may drift after model or service upgrades.
- Cross-border data flow and compliance boundaries must be reviewed in advance.
Blueprint C: Multilingual PDF delivery stack
Best fit: Teams handling papers, manuals, contracts, overseas sales assets, and multilingual branded materials
Recommended stack
- Agent: Codex or Claude Code for orchestration, batching, review, and download flows
- Delivery translation layer: Reflo
- Terminology layer: DeepL Glossary or an internal termbase
- Post-processing ecosystem: Adobe Acrobat / Adobe Translate PDF
- Quality control: PyMuPDF / qpdf / pdfcpu
Implementation steps
- Define termbases, language pairs, and document classes before letting the agent run batch orchestration.
- Route high-value files through the Reflo / DeepL / Adobe combination and reserve lighter products for lower-risk content.
- Keep a human side-by-side review step before any customer-facing delivery.
Main risks
- The closed translation layer costs more than a purely open-source stack.
- Complex PDFs still require sampled human QA.
- Errors in branded materials and contracts are expensive, so review gates remain mandatory.
Methodology
Method and evidence model
Verification date: 2026-04-15
Source types: Official product pages, official GitHub repos, help centers, dev docs, install docs
Research objects: 4 agent platforms, 36 PDF skills / tools, 6 install forms, and 3 deployable solution blueprints
Source types: Official product pages, official GitHub repos, help centers, dev docs, install docs
Research objects: 4 agent platforms, 36 PDF skills / tools, 6 install forms, and 3 deployable solution blueprints
- The evidence layer includes only official product pages, official GitHub repos, official help centers, and official developer docs. Secondary media coverage was excluded from scoring.
- Installability was divided into six forms: native skills, repo rule files, CLI/libraries, MCP, SaaS APIs, and GUI/RPA.
- Agent compatibility was judged by whether the official product exposes skills, commands, plugins, MCP, workspace files, CLIs, or APIs, not by marketing language alone.
- Public OpenAI materials now confirm both Skills and AGENTS.md for Codex, but the public install spec for native skills is still less explicit than Claude Code's. Where needed, the report marks those Codex recommendations as implementation guidance rather than full official file-by-file specification.
Sources
Official source list
To stay aligned with EEAT, this report prioritizes official domains, official GitHub repos, help centers, and official developer docs. Any inference is explicitly marked instead of being mixed with verified facts.
Codex
Claude Code
OpenClaw
Trae
Open-source PDF stack
Closed / cloud PDF stack
- OpenAI PDF files guide
- OpenAI file search
- Claude PDF support
- Claude citations
- Mistral OCR
- Mathpix PDF to Markdown
- Google Document AI overview
- Azure Document Intelligence overview
- Amazon Textract overview
- Adobe Acrobat AI Assistant
- Adobe Translate PDF
- ABBYY FineReader PDF
- Reflo upload
- DeepL file translation
- DeepL glossary for file translation
FAQ
Common questions
Is Claude Code open-source or closed-source?
As of April 15, 2026, Claude Code has an official GitHub repo. The practical classification is hybrid: an open CLI surface with proprietary models and hosted service layers.
Can Codex install PDF skills the same way Claude Code does?
Yes, but the safest publicly documented path is still AGENTS.md plus repo scripts and PDF CLI/API tools. OpenAI has publicly confirmed Skills, but the public file-system specification is still less explicit than Claude Code's.
Is OpenClaw a good fit for GUI-style PDF tools?
Yes, especially when browser, desktop, and messaging channels matter. But GUI automation is generally less stable and less auditable than CLI or API approaches.
Is Trae open-source or closed-source?
It depends on which surface you mean. Trae Agent has an official open-source MIT repo; Trae IDE and SOLO remain closer to closed hosted product surfaces.
What is the minimum viable stack for a reliable PDF agent?
Start with OCRmyPDF, Docling or MinerU, PyMuPDF or pypdf, and qpdf. Then add OpenAI, Claude, Reflo, or DeepL only where your workflow truly needs those layers.
Final recommendation
Choose install form first, PDF skill second, and model brand third
In 2026, successful PDF-agent systems are decided more by CLI/API/MCP installability, auditability, and permission design than by model branding alone. For multilingual PDF delivery, Reflo plus DeepL or Adobe is strong on the closed-source side; for local open-source baselines, OCRmyPDF, Docling, MinerU, PyMuPDF, and qpdf remain the practical core.