Scored across 26 tools
The descriptions are exceptionally explicit, with each tool naming its siblings and stating when to use which (e.g. validate_records vs validate_payment_scheme vs validate_xml_against_schema; generate_message vs its _async and _from_file variants). The only real overlap is between simulate_payment_batch and stage_payment_batch, which both run schema/scheme validation, control sums, and duplicate detection, differing mainly in the staging/token layer, so a few calls could be confused.
Every tool follows a consistent snake_case verb_noun (or verb_object) convention: list_message_types, get_input_schema, validate_records, generate_message, parse_camt053, convert_mt101, etc. Suffix modifiers (_async, _from_file) are applied predictably to the base verb, with no camelCase or mixed-style deviations.
At 26 tools the surface is heavy and sits at the top of the acceptable range, driven partly by the four-tool corpus group and the four-tool staging/clearing lifecycle. Each tool is individually justifiable given the broad ISO 20022 domain, but the set is dense enough that some consolidation (e.g. simulate vs stage) would help.
The surface covers the full payment lifecycle: discovery of message types/schemas/formats, record validation and repair, generation (memory, file, async), XSD validation, scheme rulebook checks, version migration, MT101 conversion, charset sanitisation, and parsing of pain.002/camt.053 responses, plus a market corpus with provenance and coverage metadata. No obvious dead ends remain for the stated pain/camt domain.