get_approval
ChatGPTGet the server-authored review card for one approval run returned by a pending action. On an unscoped OAuth session, pass organization with the target workspace slug or id. The card reads the canonical durable approval, exposes in-card controls only when this OAuth app context can resolve it, and always includes a review link while pending. Reading does not change workspace content or external systems. OAuth and PAT calls append a private audit/activity record.
invoke_event_action
ChatGPTAppend the declared event for a JSON-template interaction using the signed-in MCP App user as the actor. The server revalidates the source event, rendered action/value, and event-kind registry.
invoke_view_action
ChatGPTAppend the declared event for a view action using the signed-in MCP App user as the actor. The server revalidates the CURRENT view row: an action the view no longer declares is rejected.
open_view
ChatGPTOpen a Lobu view. Frame-capable hosts render it inline as an MCP App; the result also carries a web URL that renders the same view with the same params for hosts that cannot show a frame.
query_sdk
ChatGPTRun capability-scoped, read-only TypeScript through the Lobu SDK. On bare OAuth /mcp, select a workspace with await client.org(target) before workspace methods. Query entities, relationships, feeds, operations, metrics, and authorized connected-source data; write, administrative, and external-action methods are rejected by the sandbox. Use run_sdk for mutations, search_sdk to discover methods, and await ctx.sleep(ms) for bounded polling. Lobu appends a private audit/activity record for the invocation.
query_sql
ChatGPTRun a paginated, sortable, searchable read-only SQL query (member-safe). Table references auto-scope to the bound org, or pass connection to push read-only SQL fully into an external database connector. Source-backed feeds are queried explicitly with query_sdk client.feeds.readMany. Prefer client.metrics.query for declared measures; use client.query in query_sdk for simple one-shot SQL. Do NOT use positional parameters ($1, $2, …). org_slug selects a granted workspace on bare OAuth /mcp and is required when the connection has no workspace binding. The query does not change workspace content or external systems, but Lobu appends a private audit/activity record for the invocation.
resolve_approval
ChatGPTResolve the exact pending approval represented by this Lobu MCP App. This tool is app-only and requires the hidden, host-bound capability delivered with the review card.
run_sdk
ChatGPTExecute a capability-scoped Lobu SDK script. On bare OAuth /mcp, select workspace operations with await client.org(target); account discovery and conversation titles require no target. The script can create, update, or delete Lobu data and invoke connector, agent, or device operations only through documented client methods; workspace permissions, operation policies, human approvals, and audit remain enforced. Use query_sdk for reads, search_sdk to discover methods, and dry_run=true to preview write, administrative, or external calls without executing them. For connector setup, do the authorized SDK work first using connections.connect or connections.reauthenticate. When human input is required, use only exact URLs returned by tools, explain which account/app and permissions the user is authorizing, and follow the returned resume_call/completion_check. Never invent setup URLs or ask for OAuth client secrets in chat; use the returned browser setup form. Resume the requested work after verifying authorization completed.
save_memory
ChatGPTSave user-shared facts, preferences, decisions, observations, and notes to workspace memory. The returned id is immediately readable with client.knowledge.read; the result also echoes the bounded saved payload for inline display. Semantic search indexing is asynchronous and reported as indexing_status. Storage is append-only — pass supersedes_event_id to replace an existing fact (the old event is hidden from future searches without losing history). Use a stable idempotency_key when a write may be retried. Optionally attach to entities via entity_ids. Always search first to avoid duplicates.
search_memory
ChatGPTSearch local saved workspace memory: entities, facts, decisions, preferences, observations, notes, and authorized channel transcripts. To open a known memory, event, or content ID, pass query: "memory 1234" and include_content: true. entity_id is only for entity records, never memory IDs. Source-backed feeds are never queried implicitly; coverage reports local stores searched and visible source feeds with status not_queried, which agents can read explicitly through query_sdk client.feeds.readMany. Pair writes with save_memory. The search does not change workspace content or external systems. OAuth and PAT calls append a private audit/activity record.
search_sdk
ChatGPTDiscover available SDK methods and runtime helpers. Search by method name, namespace (e.g. "entities", "connections", "automations"), or keyword. Returns documentation, signatures, and access requirements for each method. (Then call methods via query_sdk for reads or run_sdk for writes. Pass mode="read" to show only query_sdk-safe methods.) The search does not change workspace content or external systems. OAuth and PAT calls append a private audit/activity record.
