Management MCP tools

Every tool takes a tenant and an environment — neither is inferred from your credential.

Start here

Tool	Purpose
`speedtrain_list_tenants`	The tenants your credential can reach. Call this first.
`speedtrain_list_environments`	The environments in a tenant. Call this second.

Most credentials reach exactly one tenant. If only one comes back, use it rather than asking which to use.

Reference

Tool	Purpose
`speedtrain_get_task_config_authoring_guide`	The complete configuration reference — field meanings, prompt assembly, rule types, gate arithmetic, worked examples.
`speedtrain_get_task_config_template`	The document skeleton to fill in, with one fully-formed variant so the shape is unambiguous.
`speedtrain_list_deployment_models`	Available deployment models and their inference settings.

Read the authoring guide before either authoring a configuration or interpreting classifier feedback. Misreading a classifier verdict is the easier of the two to get wrong — an evaluator can produce confident feedback about source data it structurally never received.

Content sources

Tool	Purpose
`speedtrain_list_indices`	Indexes in the environment.
`speedtrain_list_index_documents`	Documents in an index. Paginated.
`speedtrain_get_index_document`	One document, including its `$images`.

Task configurations

Tool	Purpose
`speedtrain_list_task_configs`	Configurations in the environment.
`speedtrain_get_task_config`	One configuration’s document.
`speedtrain_create_task_config`	Create a new configuration.
`speedtrain_push_task_config`	Push a new version of an existing configuration.
`speedtrain_list_task_config_versions`	Version history, and which one is published.
`speedtrain_activate_task_config_version`	Make a version the published one. Names the version it displaces.

Review sessions

Tool	Purpose
`speedtrain_preview_review_session`	Run a test session, including against a draft version.
`speedtrain_start_review_session`	Start a real review session.
`speedtrain_list_review_sessions`	Sessions for a configuration. Paginated.
`speedtrain_get_review_session_status`	Progress and approval rate for one session.
`speedtrain_query_review_items`	Items with their scores and feedback, filterable by decision. Paginated.
`speedtrain_get_review_item`	One item in full.

Before re-running a configuration that has run before, list its existing sessions and consider reusing one. Comparing two runs only means something if both cover the same documents.

Destinations

Tool	Purpose
`speedtrain_list_destinations`	Destinations in the environment.
`speedtrain_get_destination`	One destination, including its ingest details.
`speedtrain_create_destination`	Create a destination.
`speedtrain_set_destination_status`	Activate or deactivate it.
`speedtrain_delete_destination`	Remove it.

Pagination

Exactly three tools paginate: speedtrain_list_index_documents, speedtrain_list_review_sessions, and speedtrain_query_review_items.Each returns a total and a cursor. Pass the cursor back unchanged, with the same filters, for the next page. A null cursor is the only end signal.Every other tool returns its list complete — the server walks the platform’s pagination for you, so there is no cursor to look for.

Reading results safely

When an item’s classifier scores include a grounding or faithfulness verdict, check the rule’s source-access setting before trusting it. The item projections surface that setting next to each score, and null there means unknown, not false — read the source document before treating the verdict as authoritative. Quality checks

Overview

Authoring guide

Settings

MCP server

Start here

Reference

Content sources

Task configurations

Review sessions

Destinations

Reading results safely

​Start here

​Reference

​Content sources

​Task configurations

​Review sessions

​Destinations

​Pagination

​Reading results safely

Start here

Reference

Content sources

Task configurations

Review sessions

Destinations

Pagination

Reading results safely