Permissions and troubleshooting.

Diagnose the connection, context, operation and result independently. Start with reads and retained IDs; a new paid run or broader permission is not a generic connection test.

#Locate the failed layer before repeating work

Check connection and authentication first, then the intended Studio session and its Agent access. Inspect project and graph revisions before repairing a conflict. For an execution problem, retain and inspect the existing receipt and error. A tool missing from the client is not fixed by rerunning a node. Share cleaned identifiers and exact errors without keys or private output.

Each selected fault reveals its matching evidence: missing tools, changed tab binding, revision 7 versus 8, an original action or execution receipt, or an omitted retained-result branch. Later boundaries are not automatically diagnosed by the selected read.

#Tools are missing

Check the exact environment URL, supported MCP authentication, current client connection, scopes, account capabilities and tool profile. A server saved in client settings does not prove its tools are loaded in this conversation. Read get_mcp_guide when available. Do not invent an unavailable tool name or substitute a REST key for the MCP credential.

#The live tab cannot be used

Enable Agent access on the intended tab and inspect listed sessions. Attachments are context-bound and short-lived; reread after reconnect or a project/workspace change. An unavailable/expired attachment is not permission to continue on a headless graph. Attached project loading can be unavailable even when read, edit, or save actions are supported.

#A mutation or run is refused

Symptom	Read or check before another action
Stale editor revision	Current attachment and graph state
Stale media revision	Current media and editor revisions together
Source hash mismatch	Actual editor buffer and source hash
Saved version conflict	Current saved bundle, then reconcile
Missing required input	Actual schema and intended input value
Cost preflight unavailable	The unavailable prerequisite; no dispatch occurred
Quote expired or mismatched	Fresh bound preview and approval for changed terms

Do not replace a guard value without comparing the content it protects. A confirmation flag does not bypass quote context or establish human approval.

#An outcome is uncertain

Access: verify connection, scope, capability, profile and intended tab; do not use paid execution as a probe. Uncertain write: read the original actionId or executionId and preserve its logical key before another attempt. Incomplete display: use the matching value reader and version; absence from a bounded response does not prove an empty output.

#A result is truncated or cannot be displayed

Use the indicated exact-version graph, execution, action or capture value reader for omissions. Finish all needed pages and paths. Check whether a media URL expired independently of retained result availability. A host without Run Inspector can still use structured results.

Keep actionId, executionId, result versions, attachment handles and idempotency keys distinct in a sanitized report. Avoid exposing credentials, private prompts or source media. See Live sessions, Execution approvals, and Results for the appropriate recovery tool.