Update Transform MCP tool names and cover the extraction tools

[simoncoombes]

The server renamed three tools and added structured data extraction, so this example named tools that no longer exist and taught only half of what the server does.

Renames, verified against the live server (Unstructured Transform 0.7.2) via get_instructions and consistent with Unstructured-IO/docs#1004:

• transform_files is now start_transform_job
• check_transform_status is now check_job_status
• get_transform_results is now get_job_results

request_file_upload_url is unchanged.

The Vercel example's system prompt now also teaches extraction: parse first, keep the output_ref, draft a schema with suggest_extraction_schema_for_file if the user has not supplied one, then start_extraction_job and poll with the same status and results tools. Two things that trip agents up are called out explicitly - extraction results come back inline rather than behind a download URL, and they carry provenance that should be kept rather than reporting bare extracted_data. I raised the step budget because a parse-then-extract chain runs two polling loops, and noted that the 60s maxDuration can be tight for it.

The Cline docs here had the same gap: llms-install.md asked the agent to verify only the four parse tools, and the README described a parse only. Both now cover extraction, and the root README's tool list gains get_instructions.

I ran the full flow against production before writing it up (parse an image with strategy=vlm, suggest a schema from the resulting output_ref, extract, read back provenance-wrapped JSON) so the prompt matches real behaviour.

The install page carrying the same example code is updated in Unstructured-IO/docs#1028.

[cubic-dev-ai]

No issues found across 1 file

_{Re-trigger cubic}