- Duale AI
- Documentation
- Tools
- Tool operations
Tool operations: catalog states and call outcomes
Describes MCP catalog freshness and connection policy states, model-facing Tool naming, and how Analytics reports Tool Call outcomes.
This page covers catalog freshness, connection policy, changing MCP catalogs, model-facing Tool names, and call outcome analytics.
- Catalog freshness states and their meaning.
- Connection policy states and next steps.
- Reviewing a changing MCP catalog.
- Model-facing Tool names versus native names.
- Call outcomes and confirmed success rate.
Summaries were generated by AI.
Under Tools, Status and Catalog show whether an MCP Connection is enabled, whether its catalog is current, and which Tools Duale AI can offer. Analytics shows how MCP and SDK Tool Calls ended.
Catalog freshness
| State | What it means |
|---|---|
| Not observed | No catalog check has succeeded yet |
| Active | A recent catalog check succeeded, so compatible Tools can be offered |
| Expired | No catalog check has succeeded recently, so its Tools are not offered |
| No Tools reported | A recent catalog check succeeded and returned no Tool |
| Unavailable | The Dashboard cannot read the catalog status |
- State
- Not observed
- What it means
- No catalog check has succeeded yet
- State
- Active
- What it means
- A recent catalog check succeeded, so compatible Tools can be offered
- State
- Expired
- What it means
- No catalog check has succeeded recently, so its Tools are not offered
- State
- No Tools reported
- What it means
- A recent catalog check succeeded and returned no Tool
- State
- Unavailable
- What it means
- The Dashboard cannot read the catalog status
Saving a change to the Connection starts a catalog check at once. After you fix the server or the remote account without changing the Connection, a later check makes the catalog Active again on its own. When a check fails, Troubleshoot Tools lists what to check.
Connection policy
| State | What it means | What to do next |
|---|---|---|
| Enabled | Catalog checks run, and compatible Tools are offered after a successful check | Read the catalog freshness |
| Calls blocked | Catalog checks continue, and the Connection’s Tools are no longer offered. A call ends as Blocked without reaching the server | Review the catalog, then set Tool access to allow |
| Disabled | Catalog checks stop, and the catalog soon shows Expired. A call ends as Blocked without reaching the server. Configuration and assignments stay saved | Enable the Connection to resume checks |
- State
- Enabled
- What it means
- Catalog checks run, and compatible Tools are offered after a successful check
- What to do next
- Read the catalog freshness
- State
- Calls blocked
- What it means
- Catalog checks continue, and the Connection’s Tools are no longer offered. A call ends as Blocked without reaching the server
- What to do next
- Review the catalog, then set Tool access to
allow
- State
- Disabled
- What it means
- Catalog checks stop, and the catalog soon shows Expired. A call ends as Blocked without reaching the server. Configuration and assignments stay saved
- What to do next
- Enable the Connection to resume checks
After you delete the Connection, a call to one of its Tools ends as Failed. After you remove an Agent Identity from the Connection, a call from that Agent Identity’s Tasks ends as Denied.
Review a changing MCP catalog
The server can change its catalog at any time. While Tool access is allow, a new compatible Tool reaches every
assigned Agent Identity after the next successful check. Tool limits
states how far a catalog can grow.
Confirm MCP server compatibility states which Tools Duale AI offers to a model and how the Catalog view shows the others.
The server writes each Tool’s name, description, and input schema, which the model reads as the Tool’s definition. If you do not run the server, you do not control this text: Where text you do not control enters explains how it can steer Tool selection and arguments. What a call can do depends on the remote account’s permissions and the Connection’s Tool access setting, not on a Tool’s title or annotations.
Understand model-facing Tool names
The Catalog view shows the native Tool name that the server reports. An Agent Identity’s Tools tab, opened from the Agents list, shows the model-facing name used during a Task.
A Tool’s model-facing name can differ from its native name. It can also change during a running Task, for example when another Tool Host or Connection in your organization adds a Tool with the same name. A Tool Call always reaches the Tool the model chose, even if a later change renames it. In a Task instruction, describe the operation instead of a model-facing Tool name. Analytics lists a renamed Tool under each of its names.
Read call outcomes and reliability
Reading Analytics and the confirmed success rate requires analytics:read_metrics on tenant:<tenant_id> and a
multi-factor sign-in (aal2).
Analytics counts one final outcome for each Tool Call the model requested, except a call that a safety check blocked. The organization view shows the total, and splits it by MCP Connection or SDK Tool Host, by model-facing Tool name, and by Agent Identity.
Analytics shows the outcomes that Read a Tool Call outcome defines, and groups Not executed and any outcome without its own category as Other.
The Confirmed success rate is succeeded / (succeeded + failed): only Succeeded and Failed calls count.
When the selected period has few succeeded or failed calls, the Dashboard shows only the counts.
For one call, read the Task view. Its Effects section shows what the Task changed or sent, as far as Duale AI can tell.
On an Executor page, Success, 7 days gives the same rate for one Tool of one Agent Identity, and links to the matching Analytics view.