Skills

Drop-in skills for Claude and other agents — give your assistant document-understanding superpowers

docutray-skills v1.3.0claude plugin marketplace add docutray/docutray-skillsnpx skills add docutray/docutray-skills

docutray-skills packages the docutray agent skill, which teaches AI assistants how to use DocuTray: Claude Code, Cursor, Windsurf, Codex and 40+ other agents. In Claude Code it ships as a plugin that also bundles the DocuTray remote MCP server, so your agent can convert documents, look up document types and validate fields without any tool wiring on your side.

Why skills, not just an SDK?

LLMs work best with focused, well-described tools. A naive "here's the OpenAPI spec" approach overwhelms them. docutray-skills ships:

Hand-written prompts that explain when to call each tool.
Tight JSON schemas instead of the full OpenAPI surface.
Few-shot examples showing typical doc-understanding flows.
Error recovery patterns for low-confidence fields and retries.

The result: agents pick the right tool more often, and call it with better arguments.

Install

Pick one option. Both install the same docutray skill, so using both loads it twice.

claude plugin marketplace add docutray/docutray-skills
claude plugin install docutray@docutray

The Claude Code plugin registers the remote MCP server (https://app.docutray.com/api/mcp). Run /mcp, select plugin:docutray:docutray and sign in with your DocuTray account. No API key is needed: the server uses OAuth and you pick the organization on the consent screen. See Connect for other MCP clients.

The skill-only install does not include the MCP server. The skill then uses the DocuTray CLI, and the recommended way for agents to authenticate is docutray login --oauth, so the API key never reaches the conversation.

What the skill covers

When the DocuTray MCP tools are available, the skill uses them first. Otherwise it falls back to the CLI, the Python and Node SDKs, or the REST API.

Area	What the agent gets
Convert	Extract structured data from a document
Identify	Detect the document type automatically
Types	List, read and export extraction schemas
Steps	Run processing pipelines and check their status
Custom types	Create a new document type for your documents

See the skill overview for full install instructions, contents, and usage details.

Quick example

A Claude session with the skill installed:

User:  Pull the totals out of these three invoices and sum them.
       [attaches 3 PDFs]

Claude: I'll convert each one with DocuTray, then sum the
        net_total fields...

[converts each invoice × 3]
[reads the structured response]

Claude: The three invoices total €4,287.50. Here's the breakdown:
        - INV-1042: €1,200.00
        - INV-1043:   €987.50
        - INV-1044: €2,100.00

No glue code. The agent reads the schema, calls the tool, parses the response.

Source

docutray-skills is open source: github.com/docutray/docutray-skills