Related Servers

Each tool has a clearly distinct purpose reinforced by explicit USE WHEN / NOT FOR sections that cross-reference sibling tools (search vs filter vs find_data; get_document vs to_markdown vs extract; get_balance vs get_usage vs get_pricing). The only potential confusion is between the parallel decision-task and agent-task families, but the descriptions sharply differentiate External-mode vs Agent-stage and every verb is uniquely paired with its resource.

All 36 tools share the talonic_ prefix and overwhelmingly follow a verb_noun pattern (list_schemas, get_field, claim_decision_task, submit_agent_task, request_upload, invoke_agent_tool). The single deviation is talonic_field_values (noun-first instead of get_field_values), but it is isolated and does not undermine the otherwise highly predictable convention.

36 tools is objectively heavy, beyond the 16-25 'heavy' band. However, the platform's scope is genuinely broad — document ingestion/extraction, two kinds of task workflows (decision and agent, each with full lifecycle verbs), field registry, spec pipelines, billing, and a generic agent-tool bridge — so most tools earn their place. It feels like one consolidated platform rather than an over-split single-purpose server, but it sits at the edge of being too large to hold in working memory.

The surface covers the domain thoroughly: ingestion (request_upload/get_document/to_markdown/extract), discovery (search/filter/find_data/list_fields/get_field/field_values), pipelines (list/get/run/get_run/get_run_results), ask/answer, and two complete task lifecycles (list/claim/read-or-get/heartbeat/submit plus release/fail for decisions). Minor gaps: no document delete, no schema update/delete, and agent tasks lack the release/fail verbs their decision-task counterparts have, but none of these create dead ends for the core workflows.