Ask for the Capability, Not the Host List ​

August 6, 2026 · InfraScout Team

A dozen InfraScout tools require the target host to advertise a specific capability — LDAP search, WMI queries, ETW tracing, the systemd journal, hosting an MCP server. Until now there was no way to ask which hosts had one, so the AI listed agents and inspected them. That works exactly as long as your whole fleet fits on one page.

A capability filter ​

agent_list and agent_count both take a capability argument now, matched exactly against what each host advertises:

text
agent_count(capability="mcp_proxy")

That returns the exact number of proxy-capable hosts, with no page limit involved. The filter composes with the existing status, OS-family, search, and group filters, and works on agent_list when you want the hosts themselves.

Matching is exact rather than partial on purpose — a partial match for etw would also select hosts that only advertise etw_autologger.

A page is not the fleet ​

Agent listings have always reported an honest total: the count is never capped, only the page is. But nothing told the AI what that obliged, so a model that fetched one page and found no domain controller could reasonably conclude the fleet had none. On forty hosts that conclusion is right. On four hundred it is a silent wrong answer — worse than an error, because nothing about it looks wrong.

Both tools now state it outright: a result carrying has_more: true can never be used to conclude absence. To answer "does any host have X", pass the filter that expresses X and read the total.

Host inventory past 100 systems ​

The same problem had a concrete instance in the inventory tools. inventory_list_hosts hit a hundred-row limit it did not report: real inventoried hosts past the hundredth came back marked not inventoried, and a filtered call reported a total of a hundred or fewer with pages beyond it unreachable.

Fixed — the listing pages through properly and the reported total is the true match count. If you have more than a hundred inventoried systems and have been looking at coverage numbers, they were understated.

A host picker that stopped dropping candidates ​

The portal's picker for choosing which agent hosts an MCP server asked for more agents than the listing would return, so on a fleet past two hundred hosts eligible candidates were missing from the dropdown with nothing to indicate the list was partial. It now filters by the hosting capability directly. A host already assigned but since stripped of the capability is merged back in, so it keeps its name and settings rather than vanishing from its own configuration.

Alongside this, the agents, insights, and executions lists coalesce rapid updates instead of refetching on every event, so they stay responsive on a large fleet while things are actively happening.

See Agent Tools for the filter arguments.

Questions or feedback? Reach us at info@infrascout.cloud.