The MCP tool directory.

Add product to one or more categories. Idempotent — already-assigned categories are no-ops. Response: { productId, categoryIds, added: count }. Errors: { error: { code: 'not_found', ... } }.

Remove product from one or more categories. Idempotent — non-assigned categories are no-ops. Response: { productId, categoryIds, removed: count }. Errors: { error: { code: 'not_found', ... } }.

Publish product on one or more channels. Channel codes must already exist for the tenant (otherwise the underlying invariant rejects the write). Idempotent. Response: { productId, channelCodes, added: count }. Errors: { error: { code: 'not_found' | 'unprocessable', ... } }.

Unpublish product from one or more channels. Idempotent. Response: { productId, channelCodes, removed: count }. Errors: { error: { code: 'not_found', ... } }.

Create or upsert a single product by (sourceSystem, externalId). Re-running the same call returns the same productId and updates the existing record — never creates a duplicate. For ad-hoc creates with no upstream system, pass sourceSystem='mcp' (or another stable string the agent owns) so subsequent runs from the same agent stay idempotent. Response: { productId, sourceSystem, externalId, action: 'created'|'updated' }. Errors: { error: { code: 'not_found' | 'bad_input', ... } }.

Get a single product with its variants and resolved attribute values. Use the optional `fields` whitelist to skip heavy sections when payload size matters. Allowed `fields` values: product, attributes, variants, channelStatuses. Attribute values are returned in their native JSON shape (string/number/bool/{lang:value}/["..."]) matching the PimValue wire format used by the HTTP API. Each attribute carries its definition code + dataType so the agent can interpret the value without a separate lookup. Response: { product?: {...}, attributes?: [{ attributeDefinitionId, code, dataType, scope, unit, value, updatedAt }], variants?: [{ id, sku, name, ean, gtin, attributes: [...] }], channelStatuses?: [...] }. Errors: { error: { code: 'not_found', ... } }.

Search products in the current tenant. At most one filter dimension is honored at a time (in priority order: query > schemaId > productTypeId > categoryId > channelCode > traitId). If none are set, returns the tenant's products in default order. Pagination: 1-indexed `page` + `pageSize` (max 50). Response: { products: [{ id, schemaId, primaryTypeId, traitIds, categoryIds, channelCodes, skuList, productNumber, status, purpose, sourceSystem, sourceSystemProductId, updatedAt }], page, pageSize, totalCount, totalPages }. Errors: { error: { code: 'bad_input', ... } }.

Set the product's lifecycle status. Allowed values: draft, active, archived, requiresAttention. Use this for 'soft delete' (status=archived); there is no hard delete via MCP in v1. Response: { productId, status }. Errors: { error: { code: 'not_found', ... } }.

Assign one or more traits to a product. Traits add their attribute definitions to the product's editable surface. Idempotent. Response: { productId, traitIds, added: count }. Errors: { error: { code: 'not_found', ... } }.

Unassign one or more traits from a product. Note: existing attribute values for the trait's attributes are NOT deleted automatically — call pim.products.attributes.set with mode=replace afterwards if you want to drop them. Idempotent. Response: { productId, traitIds, removed: count }. Errors: { error: { code: 'not_found', ... } }.

Create a product type in a schema, or return the existing one if `code` already exists in the schema. ProductType is the flat product classification within a schema (NOT category/sales nav). Response: { productType: {...}, created: bool }. Errors: { error: { code: 'not_found' | 'bad_input', ... } }.

List product types in a schema. ProductTypes are the flat product classification within a schema (NOT to be confused with categories, which are sales navigation). Response: { productTypes: [{ id, schemaId, code, labels, defaultTraitIds, parentCode, level, isActive, productCount }] }. Errors: { error: { code: 'not_found', ... } }.

Create a tenant-owned PIM schema, or return the existing one if `name` already exists for this tenant. Only Master and Recipient kinds are creatable via MCP — Standard (eClass/ETIM/UNSPSC) and Connector schemas are platform-managed. Response: { schema: {...}, created: bool }. Errors: { error: { code: 'bad_input', ... } }.

Get a PIM schema by id, with productTypes, traits, attributeDefinitions, and channels embedded. Use the optional `fields` param to limit the response to specific top-level keys when the full payload is too large. Allowed `fields` values: schema, productTypes, traits, attributeDefinitions, channels. Response: { schema?: {...}, productTypes?: [...], traits?: [...], attributeDefinitions?: [...], channels?: [...] }. Errors: { error: { code: 'not_found' | 'forbidden', ... } }.

List PIM schemas in the current tenant. By default returns only tenant-owned, editable schemas (Master + Recipient). To inspect Import (file-import) or Standard (eClass/ETIM/UNSPSC) or Connector schemas, pass `kind` explicitly. Response: { schemas: [{ id, name, kind, contentLanguage, productTypeCount, traitCount, categoryCount, attributeDefinitionCount, isReadOnly, updatedAt }] }. Errors: { error: { code: 'bad_input', ... } }.

Set the schema's default LLM role/context (the system-prompt-style preamble used in AI ops). Pass null to clear. Response: { schemaId, defaultAiRole }. Errors: { error: { code: 'not_found' | 'bad_input', ... } }.

Set the schema's default channel codes — the codes auto-assigned to new products in this schema. This replaces the existing list. Pass [] to clear. Response: { schemaId, defaultChannelCodes }. Errors: { error: { code: 'not_found', ... } }.

Set the schema's product-number template. C = category code character, N = sequential digit, any other character = literal. Example: 'CCCC-NNNNNNNN' renders 'PUMP-00000001'. Pass null to clear. Response: { schemaId, productNumberTemplate }. Errors: { error: { code: 'not_found', ... } }.

Bind one or more attribute definitions to a trait by setting their `traitId`. Each attribute definition in the list has its `traitId` updated to the given trait. Idempotent — already-bound definitions are no-ops. Definitions in other schemas or other tenants are skipped with a reason. Response: { traitId, attributeDefinitionIds, bound: count, skipped: [{ id, reason }] }. Errors: { error: { code: 'not_found', ... } }.

Unbind attribute definitions from a trait — clears their `traitId` only when it equals the given trait. Definitions bound to other traits are left untouched. Idempotent. Response: { traitId, attributeDefinitionIds, unbound: count, skipped: [{ id, reason }] }. Errors: { error: { code: 'not_found', ... } }.

Create a trait in a schema, or return the existing one if `code` already exists. Traits are reusable groups of attribute definitions; products opt in to traits to gain their attributes. To bind attribute definitions to the trait after creation, use pim.traits.attributes.add (or set `traitId` directly when creating attribute definitions). Response: { trait: {...}, created: bool }. Errors: { error: { code: 'not_found' | 'bad_input', ... } }.

Get a trait by id, with its trait-scoped attribute definitions resolved. Response: { trait: {...}, attributeDefinitions: [...] }. Errors: { error: { code: 'not_found', ... } }.

List traits in a schema. Traits are reusable groups of attribute definitions; products opt in to traits to gain their attributes. Response: { traits: [{ id, schemaId, code, labels, sortOrder, isActive }] }. Errors: { error: { code: 'not_found', ... } }.

Runs a single real LLM inference directly against a model (no flow), and returns the model's response plus token usage and cost. This EXECUTES a billed provider call and consumes tenant credits. Useful for testing a model/prompt, comparing reasoning-effort levels, or reproducing behavior. The returned runId can be passed to get_run for the full per-iteration detail. For server-side tool execution and multi-step agents, run a flow instead.

Updates a template's prompt text, model, response schema, and/or tool assignments in the active or latest version. If the version is locked, automatically forks it first (the fork is a draft — activate it via the UI or API). Falls back to the latest version if no version is activated. Returns confirmation with version ID and whether a fork was created.