Skip to main content
17 core tools are always available. 4 tools require the Google Cloud runner.

​
Environment

List configured environments and current environmentMCP tool env_listInputThis tool takes no input.
Get the current environment for this MCP sessionMCP tool env_currentInputThis tool takes no input.
Switch active environment and reset session authenticationMCP tool env_switchInput
​
string
required
Target environment name

​
Workspace

Get the current workspace slug for this MCP sessionMCP tool tenant_currentInputThis tool takes no input.
Set workspace slug for session-scoped X-Tenant headerMCP tool tenant_setInput
​
string
required
Workspace slug for X-Tenant header
Clear workspace slug used for session-scoped X-Tenant headerMCP tool tenant_clearInputThis tool takes no input.

​
Discovery

Search permitted Workspace or Admin Parable-authoring tools without loading their input schemasMCP tool tool_searchInput
​
integer
Maximum results from 1 to 100
​
string
required
Keyword search for tool title, handle, namespace, capability, or description
Enable one exact permitted Workspace or Admin Parable-authoring tool for this MCP sessionMCP tool tool_enableInput
​
string
required
Exact handle returned by tool_search
Remove one previously enabled Workspace or Admin Parable-authoring tool from this MCP sessionMCP tool tool_disableInput
​
string
required
Exact enabled Workspace or Admin Parable-authoring tool handle

​
Authentication

Return auth status for the current MCP sessionMCP tool auth_statusInputThis tool takes no input.
Clear any session token stored by this MCP serverMCP tool auth_logoutInputThis tool takes no input.
Consume a single-use auth token (magic-link, signup, or invite) and store session token internallyMCP tool auth_consume_auth_tokenInput
​
string
required
Single-use auth token (UUID)

​
Health

Get a comprehensive health snapshot for a workspace: connector statuses, sync freshness, infrastructure metadata, and GCS data presence. Accepts workspace UUID or slug (auto-resolved). Aggregates multiple admin API calls into a single response.MCP tool tenant_healthInput
​
string
required
Workspace UUID or slug (e.g. ‘nadim-obs’ or ‘4G0Io2KzeWwoMkECk957ZM’)
Get connector operational status for a specific workspace. Returns summary counts (healthy, needs attention, errored) and a list of connectors with issues. Accepts workspace UUID or slug (auto-resolved). For per-connector sync freshness, use tenant_health instead.MCP tool pipeline_statusInput
​
string
required
Workspace UUID or slug (e.g. ‘nadim-obs’ or ‘4G0Io2KzeWwoMkECk957ZM’)

​
Google Cloud

Execute a read-only gcloud CLI command and return parsed JSON output. Pass the subcommand and arguments without the leading ‘gcloud’. —format=json is appended automatically. Only read-only commands are allowed (logging read, traces describe, storage ls, etc).MCP tool gcloud_readRequires Google Cloud runnerInput
​
string
required
gcloud subcommand and arguments, e.g. ‘logging read “severity>=ERROR” —project=parable-prod —limit=50’
Get the GCP project ID for the current session environmentMCP tool gcp_projectRequires Google Cloud runnerInputThis tool takes no input.

​
Alerts

List all configured platform alerts with their IDs, display names, and affected services. Use this to discover alert IDs before calling alert_triage.MCP tool alert_listRequires Google Cloud runnerInput
​
string
Optional: filter alerts by service name (e.g. ‘web-api’, ‘gke’, ‘ingestion-worker’)
Triage a platform alert by ID: fetch the runbook, execute relevant log queries, and return console deep-links. Call alert_list first to discover valid alert IDs.MCP tool alert_triageRequires Google Cloud runnerInput
​
string
required
Alert ID from the catalog (e.g. ‘ingestion-flow-failure’, ‘pod-crashloop-backoff’)

​
Ponder

Send one Message on a Problem and follow Ponder’s reply as it streams. Progress notifications report each tool Ponder uses and the reply text so far; the result is the finished user and assistant Messages. The reply runs on the server and every open page sees it live, so if this call is cut off, call await_ponder_reply. To end a reply early, pass the result’s assistantMessageId to stop_ponder_reply. A reply that needs a tool approval stops until a person approves it in Ponder; the result’s waitingForApproval names the tools.MCP tool send_ponder_messageInput
​
object
Optional orientation, the same current-view object a browser sends; never a write grant
​
string
Optional UUID for this Message; resending the same id never starts a second reply
​
string
workspace (default) or admin; admin needs Admin Ponder permission and acts on Admin Parables
​
string
Persona id. Omit to use the Problem’s assigned Persona, then your last selection, then Automatic
​
string
required
Problem (thread) id
​
string[]
Power (Agent Skill) ids to apply to this Message
​
string
Profile: simple, standard (default), or strong
​
string
required
Message text (Markdown)
Wait until a Ponder reply on a Problem finishes, then return it. Waits on the Problem’s realtime channel, not by polling, and sends a progress notification whenever the thread changes. Identify the reply by assistantMessageId, or by messageId (the user Message it answers). Returns at once when the reply is waiting for a person to approve a tool.MCP tool await_ponder_replyInput
​
string
The assistantMessageId a send_ponder_message or await_ponder_reply result returned
​
string
User Message id (the id sent to send_ponder_message)
​
string
required
Problem (thread) id
​
number
Longest wait in seconds (default 600, at most 3600)