Skip to content

MCP Tools

Tools exposed by the Studio MCP server. The last column is the token scope and role a tool needs; OAuth sessions get both scopes.

Tool What it does Access
search_knowledge_base Full-text search over the generated knowledge of datasets and storages Datasets: read
list_datasets List datasets that have a complete version, filtered by namespace and project Datasets: read
get_dataset Return a dataset's schema, versions, knowledge page, and up to 5 preview rows Datasets: read
get_row_content Read the file a dataset row points to, as an image or text Datasets: read
list_storages List indexed storage roots Datasets: read
get_storage Return an indexed storage's listing statistics and knowledge page Datasets: read
get_knowledge_base Return a Markdown index of enriched datasets and storages Datasets: read
enrich_dataset Generate the knowledge page for one dataset, using the team's provider and budget Datasets: write
enrich_all Generate pages for every dataset and storage that lacks a current one Datasets: admin
get_datachain_guide Return the DataChain SDK rules a run_job script must follow Jobs: read
list_clusters List the team's compute clusters, default first Jobs: read
run_job Submit a Python DataChain script as a Studio job Jobs: write
get_job Return a job's settings, status, progress, errors, and produced dataset versions Jobs: read
get_job_logs Return a job's most recent log lines Jobs: read
cancel_job Cancel an active job Jobs: write

Limits

  • get_dataset previews at most 5 rows and truncates each cell to 200 characters.
  • get_row_content reads files from s3://, gs://, az://, and hf:// only, up to 4 MiB; JPG, PNG, GIF, and WebP come back as images, UTF-8 files as text.
  • search_knowledge_base returns at most 50 matches per call; list_datasets at most 200 datasets per page.
  • Requests are rate limited per user and per IP address; a limited request gets HTTP
  • Wait and retry.
  • Knowledge search and the knowledge index cover only datasets and storages with a generated page; enrich_dataset and enrich_all generate pages using the team's provider and budget.

Example prompts

What datasets do we have about product images? Show the schema of the largest one.

Find storages with audio files and summarize what the knowledge base says about them.

The loop that makes an agent and Studio worth combining: a question comes in, no existing column answers it, the agent runs a pass over the files, saves the result as a dataset, and the next person who asks gets it from the store.

Do we already have embeddings for the oxford-pets images?

Which pets in s3://dc-readme/oxford-pets-micro/ were photographed outdoors?

Why did job <job-id> fail? Show the relevant log lines.