Platform actions for Trace AI

Trace AI can perform read and write operations on your tenant data through MCP tools registered in mcp_resources. Tools use the same tenant and user context as the console UI.

Activation

There is no global “platform actions” feature flag. Each tool is enabled individually via mcp_resources.is_active. Disable a tool by setting is_active = 0 on its row.

Skills + tools workflow

See Platform skills quick reference for the full skill → tool map.

Load the platform-actions skill (load_platform_skill) for orchestration guidance.
For writes, call describe_* tools first to learn accepted fields.
Call list_* / get_* read tools to gather ids and current state.
Use request_human_input (confirm) before destructive or high-risk actions when appropriate.
Call the write or execute tool with a minimal payload.

Read tools (Phase 1)

Tool	Purpose
`list_entities`, `get_entity`	Entity catalog
`list_products`, `get_product`	Products
`list_contracts`, `get_contract`	Contracts
`list_processes`, `get_process`	Process definitions
`list_integrations`, `get_integration`	Integrations (no secrets in list)
`list_webapps`, `get_webapp`	WebApps
`query_datapool`	Datapool rows by schema
`query_monitor_logs`	Process/integration execution logs
`list_notifications`	User notifications
`read_todo_list`	Tenant chat TODO list

Describe tools

Tool	Purpose
`describe_entity`	Entity create/update field contract
`describe_product`	Product create/update contract
`describe_contract`	Contract create/update contract
`describe_process`	Execute requirements, triggers, `executable_by_ai_agents`
`describe_tenant_integration`	Integration operations and payload shapes

Write tools (Phase 2)

Tool	Notes
`create_entity`, `update_entity`, `delete_entity`	Delete archives (`status=archived`)
`create_product`, `update_product`, `delete_product`	Delete sets `status=DELETED`
`create_contract`, `update_contract`, `delete_contract`	Hard delete on contract row
`execute_process`	Requires Executable by AI Agents on the process
`run_process_trigger`	Fire trigger; same opt-in gate as `execute_process`
`test_integration`	Configuration validation (same as UI test)
`execute_tenant_integration`	Requires Executable by AI Agents on the integration
`create_integration`, `update_integration`, `delete_integration`	Integration CRUD
`create_datapool_schema`, `update_datapool_schema`	Datapool schema management
`insert_datapool_rows`, `update_datapool_rows`, `delete_datapool_rows`	Datapool row CRUD
`create_entity_relationship`, `delete_entity_relationship`	Entity linkage graph

Builder tools (Phase 3)

Tool	Notes
`create_process`, `update_process`, `retire_process`	Process definitions; `retire_process` sets inactive + `<retired>` name prefix
`create_process_step`, `update_process_step`, `delete_process_step`	Step CRUD with compile validation on code snippets
`create_process_trigger`, `update_process_trigger`, `delete_process_trigger`	Trigger management
`create_webapp`, `update_webapp`, `create_webapp_version`, `publish_webapp`, `unpublish_webapp`, `delete_webapp`	WebApp lifecycle and `page_content`

Autonomic Agent tools

Load load_platform_skill with skill_id: autonomic-agents. These tools require a tenant-admin role, AUTONOMIC_AGENTS_ENABLED=1, and a Trace AI caller:

Tool	Notes
`describe_autonomic_agent`, `list_autonomic_agents`, `get_autonomic_agent`	Current definition contract and discovery
`create_autonomic_agent`, `update_autonomic_agent`	Save definition and compile its owned wake ProcessFlow
`disable_autonomic_agent`, `delete_autonomic_agent`	Confirm first; delete fails while runs are in flight
`run_autonomic_agent`	Confirm first; synchronous manual test turn
`list_autonomic_agent_runs`, `get_autonomic_agent_run`	Run status, heartbeat, decision, and errors

Autonomic Agents cannot load these management tools to create or run peers recursively.

Trace AI subagent tools

Load load_platform_skill with skill_id: subagents. The tools are restricted to Trace AI tenant administrators and scope run inspection/cancellation to the current parent Trace session.

Tool	Notes
`list_subagent_catalog`	Active definitions with `delegatable_by_trace` enabled
`delegate_subagent`	Queue isolated work and return `run_id`, `queue_id`, and `execution_id`
`get_subagent_run`, `list_subagent_runs`	Run state, final decision, and ordered progress events
`cancel_subagent_run`	Immediate for queued work; cooperative at safe boundaries for running work

See Trace AI subagents.

Platform code validation tools

Load load_platform_skill with skill_id: platform-code-generation.

Tool	Notes
`describe_process_step_code_contract`	Supported languages, execution modes, capabilities, and restrictions
`validate_process_step_code`	Read-only sandbox/security and JavaScript/TypeScript compile/syntax validation
`describe_webapp_code_contract`	WebApp browser/content architecture and restrictions
`validate_webapp_content`	Read-only HTML, CSS, JavaScript, and HTTP-header validation

Validation does not create, update, publish, or execute a platform resource.

Platform admin tools (Phase 4)

Tool	Notes
`create_platform`, `update_platform`, `update_platform_strategy`	Business platform CRUD; strategy uses section prefixes from platform strategy config
`upload_document`, `move_document`, `create_document_review`	Tenant document filesystem + review workflow
`mark_notification_read`, `delete_notification`	Scoped to current user notifications
`upload_platform_file`, `delete_platform_file`	Platform-scoped file namespace (`content_base64` for uploads)

Opt-in flags

Resource	UI setting	MCP gate
Integration execute	Integration → Executable by AI Agents	`execute_tenant_integration`
Process execute	Process edit → Executable by AI Agents	`execute_process`, `run_process_trigger`

Default for both flags is off. Agents cannot set these flags via MCP create/update tools; ask the user to toggle them in the console UI.

Human in the loop

Destructive writes and execute_process should be preceded by request_human_input with kind: confirm in production workflows. The agent receives an explicit approved/denied result in the tool loop.