Loading skill
Install any skill in seconds. Free to start, no credit card required.
Get Started Free →automateyournetwork/multivendor-fleet-ops
Ask one question of many devices at once — across mixed vendors — and get per-device answers.
Passed 15 of 19 test cases with the skill, and 7 of 19 without it.
and 10 more in the full report
9 fixed8 passed either way5 still failing0 broke
Tokens, turns and time describe resource use, not a dollar price.
One of 9 test cases the skill fixed. Same model, same prompt, with and without the skill.
If we target 12 routers in a fleet query and 2 of them are offline due to a WAN outage, how many device result objects will be present in the returned execution payload?
| Checked | Without skill | With skill |
|---|---|---|
| States exact returned count is 12. | Passed | Passed |
| States requested count always equals returned count. | Missed | Passed |
The instructions your model reads when the skill loads.
Ask one question of many devices at once — across mixed vendors — and get per-device answers.
MCP server: multivendor-cli-mcp · Tools: run_fleet, get_facts
run_fleettarget is a group name, a comma-separated device list, or all. Supply exactly one of:
command — raw CLI, only for platforms this server ownsgetters — NAPALM normalized facts, permitted read-only on any platformrun_fleet(target="edge", getters=["get_facts","get_interfaces"])
run_fleet(target="srl1,srl2", command="show version")
requested always equals returned. Every targeted device appears, including failures. A
silently absent device would read as success, which is the most dangerous possible output for a fleet
query — you'd scan the list and see no problem where there is one.
summary counts by status. One device failing never aborts the others.
Defaults: 10 concurrent workers, 30s per-device timeout, both overridable. Ten rather than Nornir's own 20 because each worker holds an SSH session and devices commonly cap concurrent management sessions at 5–15.
get_facts returns one shape across vendors, which no other NetClaw server can do. pyATS and
junos-mcp each answer well for their own platform, but their shapes differ, so a question spanning
vendors otherwise needs reconciling by hand.
Where a platform has no NAPALM driver — SR Linux, FRR, VyOS, MikroTik — you get a row with
available: false and a gap_reason, not a missing row. That distinction is deliberate: "this
platform cannot tell us" and "there are none" are different answers, and conflating them produces a
wrong one. provenance is never faked as napalm for scraped output.
test cases passed, vs 7 / 19 without
More skills from the same source and nearby in the registry, with their headline benchmark lift.