docyrus-architect（多赛勒斯建筑师）

Docyrus Architect

Guide for using docyrus-architect MCP tools to manage and query data sources in Docyrus.

Tool Overview

Discovery Tools

• get_apps — List tenant apps. Use before create_data_source to find the target tenantAppId.
• get_data_source_list — Search data sources by name/description or app ID.
• get_data_source_list_with_fields — Same as above but includes field names and types.
• get_data_source_metadata — Get full metadata (fields with IDs, types, slugs, enums, relations) for a data source. Always call this before querying to discover field slugs and relation targets.
• get_enums_by_field_id — Get enum options for select/status/tagSelect fields.
• read_current_user / read_tenant_user — Get user info.

Data Source CRUD

• create_data_source — Create a new data source (table). Default fields auto-created: id, autonumber_id, name, record_owner, created_on, created_by, last_modified_by, last_modified_on.
• create_data_sources_batch — Create multiple data sources with fields and enum options in a single atomic transaction. Supports cross-referencing data sources within the batch by slug for field-relation types. Max 20 data sources per batch.
• update_data_source — Update data source properties.
• delete_data_source — Delete a data source and all its data.

Field Management

• create_fields — Batch create fields. Set relationDataSourceId for field-relation types.
• update_fields — Batch update fields. Non-CUSTOM fields get customization records.
• delete_fields — Batch delete fields by ID.

Enum Management

• create_enums — Create enum options for select/tagSelect/status fields. Pass fieldId for field-specific enums or enumSetId for shared enum sets.
• update_enums — Update enum option name/slug/color/icon.
• delete_enums — Delete enum options.

OpenAPI Spec

• regenerate_openapi_spec — Regenerate and upload the tenant's OpenAPI spec after data source or field changes. Accepts optional dataSourceSlugs (string array) to limit scope; omit to include all data sources. Returns the publicUrl of the uploaded spec. Call this after any create_data_source, update_data_source, delete_data_source, create_fields, update_fields, or delete_fields operation to keep the spec in sync.

Query & Compute

• query_data_source — Read data with filtering, sorting, aggregation, formulas, pivots, child queries. See references/data-source-query-guide.md for complete query syntax.
• evaluate_jsonata — Test JSONata expressions. Use for validating computed field formulas.

Custom Queries

• get_custom_queries — List non-archived custom queries for the tenant.
• get_custom_query_by_id — Read full custom query definition (name, description, query, filters).
• create_custom_query — Create a saved SQL template using Handlebars variables and optional filter definitions.
• update_custom_query — Update selected fields of an existing custom query (partial update).
• delete_custom_query — Soft-delete (archive) a custom query.
• run_custom_query — Execute a saved custom query with runtime filters, pagination offset, and optional simulate mode.

Common Workflows

Create a Data Source with Fields and Enums

Single data source approach:

1. Call get_apps to find the target app ID
2. Call create_data_source with title (plural), name (singular), slug (singular snake_case)
3. Call create_fields with all custom fields (default fields already exist)
4. For select/tagSelect/status fields, call create_enums with the field ID from step 3
5. Call regenerate_openapi_spec to update the OpenAPI spec

Batch approach (recommended for multiple data sources):

1. Call get_apps to find the target app ID (optional, defaults to "Default Customizations App")
2. Call create_data_sources_batch with an array of data source definitions (max 20)

- Each data source includes title, name, slug, and fields array - Fields can include inline enumOptions for select/status/tagSelect fields - Use slug references for relationDataSourceId to reference other data sources in the same batch - All data sources, fields, and enums are created in a single atomic transaction

1. Call regenerate_openapi_spec to update the OpenAPI spec

Query Data

1. Call get_data_source_metadata to discover field slugs, types, and relations
2. Call query_data_source with appropriate columns, filters, and sorting
3. For advanced queries (aggregations, formulas, pivots, child queries), read references/data-source-query-guide.md

Modify Existing Data Source

1. Call get_data_source_metadata to see current fields
2. Use create_fields / update_fields / delete_fields as needed
3. For enum changes, use get_enums_by_field_id first, then create_enums / update_enums / delete_enums
4. Call regenerate_openapi_spec to update the OpenAPI spec

Manage a Custom Query Lifecycle

1. Call get_custom_queries to find an existing query or confirm naming
2. Call create_custom_query with name, query, and optional description/filters
3. Call get_custom_query_by_id to verify the saved template and filter definitions
4. Use update_custom_query for targeted edits (rename, revise SQL template, update filter definitions)
5. Use delete_custom_query only when archival is explicitly requested

Run and Debug a Custom Query

1. Call get_custom_query_by_id first to inspect SQL template and available filter slugs
2. Build runtime filters with rules[].field values that match the query's filter slugs
3. Call run_custom_query with simulate: true for complex or untrusted queries to inspect plan output
4. Call run_custom_query with simulate: false (or omitted) to fetch real data
5. Use offset for pagination and expect a max of 50,000 rows per execution

Key Rules

Data Source Creation

• title is plural (e.g., "Sales Orders"), name is singular (e.g., "Sales Order"), slug is singular snake_case (e.g., "sales_order")
• Use defaultEditFormTarget: "tab" for complex forms, "side" for simple ones
• Enable pluginActivityView for CRM-type data sources (leads, contacts, deals)
• Enable pluginComments for collaborative data sources
• Enable pluginFile when users need to attach files to records
• Enable pluginDocyment when users need rich text documents per record

Field Types

• field-relation requires relationDataSourceId — the ID (UUID) of the related data source, or in batch operations, the slug of another data source in the same batch
• field-list is a virtual field showing child records (one-to-many) — not stored in DB
• field-select / field-tagSelect / field-status need enum options:

- In create_fields: call create_enums after field creation - In create_data_sources_batch: include inline enumOptions array in the field definition

• field-formula uses JSONata expressions — test with evaluate_jsonata first
• field-inlineData stores array of objects, field-inlineForm stores single nested object
• Field slug must be snake_case matching ^[a-z][a-z0-9_]*$
• Enum option slug must be snake_case matching ^[a-z][a-z0-9_]*$ (auto-generated from name if not provided)

Batch Operations

• create_data_sources_batch creates up to 20 data sources in a single transaction
• Within a batch, reference other data sources using their slug for relationDataSourceId
• For external data sources (created outside the batch), use the UUID for relationDataSourceId
• Data source slugs must be unique within the batch (validation error otherwise)
• Inline enum options are defined per-field using the enumOptions array
• Each enum option in enumOptions can specify: name, slug, color, icon, sortOrder
• If a batch operation fails, all changes are rolled back (atomic transaction)
• Use batch operations when creating related data sources to avoid multiple API calls and ensure consistency

Querying

• Use dataSourceId (UUID) to identify which data source to query
• columns is a comma-separated string of field slugs, not an array
• For aggregations, always use id field for count calculations
• Relation expansion: relation_field(sub_field1, sub_field2) selects nested columns
• Spread operator: ...relation_field(alias:sub_field) flattens into root object
• Filter on related fields: rel_{{relation_slug}}/{{field_slug}}
• Date filters have shortcut operators like today, this_month, last_30_days

query_data_source Required Parameters

All parameters are required in the MCP tool schema (most accept null):

• dataSourceId: string (required, non-null)
• columns: string | null
• filters: object | null
• filterKeyword: string | null
• orderBy: array | null
• limit: number | null (default: 1000)
• offset: number | null
• fullCount: boolean | null
• recordId: string | null (fetch single record by ID)
• calculations: array | null
• distinctColumns: array | null
• formulas: array | null
• childQueries: array | null
• pivot: object | null

Custom Query Rules

• Treat custom query query content as a Handlebars SQL template, not static SQL.
• Use built-in context variables in templates when relevant: TENANT_ID, TENANT_SCHEMA, USER_ID, USER_EMAIL, USER_FIRSTNAME, USER_LASTNAME, USER_FULLNAME.
• Use {{filter FILTERS.<slug> <column_expression>}} for optional runtime filtering. If no runtime value is provided, the helper resolves to 1=1.
• Keep filters definitions in create_custom_query / update_custom_query aligned with template usage:

- template usage: FILTERS.<slug> - runtime rule field: <slug> - filter definition slug: <slug>

• Prefer simulate: true before running expensive queries to inspect EXPLAIN (ANALYZE, BUFFERS, FORMAT JSON) output.
• Respect runtime limits on run_custom_query: 50,000 row cap, 15s timeout for normal execution, 30s timeout for simulate mode.
• Use delete_custom_query as an archival action (soft delete), not permanent deletion.
• For JSONB-backed simple data sources, access custom fields with (table_alias."data"->>'field-uuid')::type.

References

• Data Source Query Guide — Up-to-date reference for query_data_source including columns, filters, orderBy, pagination, calculations, formulas, pivots, child queries, and operator details.
• Formula Design Guide (LLM) — Up-to-date guide for designing formula payloads used in query requests.
• Custom Query Guide — Full lifecycle and execution reference for get_custom_queries, get_custom_query_by_id, create_custom_query, update_custom_query, delete_custom_query, and run_custom_query.