Meraki Monitoring & Diagnostics (read-only)

MCP Server

• Server: Cisco's official Meraki MCP — developer.cisco.com
• Endpoint: https://mcp.meraki.com/mcp (remote HTTP, no local install)
• Self-host fallback: CiscoDevNet/cisco-meraki-mcp-official (Apache-2.0)
• Requires: MERAKI_DASHBOARD_API_KEY — use a read-only dashboard key
• Tools: exactly two — semantic_search (discover) and execute_api (invoke)

Read-only, structurally

All 431 mutating Meraki operations (174 POST, 186 PUT, 71 DELETE) are absent from the capability catalogue. Only non-deprecated GETs are built in, so updateNetwork, rebootDevice, blinkDeviceLeds and friends return Capability not found.

Do not attempt writes and do not offer them. There is no ServiceNow CR path here because there is nothing to gate — a change must be made in the Meraki dashboard directly.

How to call it

Two capability IDs may be called directly; everything else must be discovered:

1. execute_api → getOrganizations — find accessible org IDs
2. execute_api → getOrganizationNetworks — find network IDs in the chosen org
3. semantic_search → describe the intent in words → returns ranked capability_ids
4. execute_api → the chosen capability_id + its required parameters

Discover, do not guess. An earlier version of this skill hardcoded 80 method names and 54 of them did not exist in the Meraki API at all — they failed regardless of server. semantic_search is the guard against that: it can only return IDs that exist.

Reading results honestly

• One page only. execute_api returns a single page. Request a bounded page size when supported, and never present a page as the complete dataset.

• Empty is not absent. n=0 means this network reported none, never none exist. Live proof from a real sandbox org: getOrganizationDevices returns 0 devices while the same org has a fully configured network with 15 SSIDs.

• Three distinct errors, three different fixes:

  | Error | Meaning | Fix | |---|---|---| | Capability not found | ID is mutating, deprecated, or invented | semantic_search for a real one | | Resource not found | valid ID, wrong org/network/serial | re-derive the ID from step 1–2 | | Invalid parameters | valid ID, missing a required parameter | read the capability's parameters |

  Never report any of the three as "no data" — none of them mean that.

Verified capabilities

Live-verified against a real org:

| Capability ID | Verified result | |---|---| | getOrganizationConfigurationChanges | 3 change records | | getOrganizationApiRequestsOverview | API request analytics present | | getOrganizationLicensesOverview | licensing (co-term) present | | getOrganizationDevices | 0 devices | | getOrganizationInventoryDevices | 0 inventory | | getOrganizationDevicesStatuses | Capability not found — deprecated upstream |

That last row matters: 8 deprecated GETs are filtered out of the catalogue. A Capability not found on a name you remember working means it was deprecated, not that the data is gone — semantic_search for the current equivalent.

There are 56 real monitoring capabilities, including getDeviceLossAndLatencyHistory, getOrganizationDevicesAvailabilities, getOrganizationDevicesUplinksLossAndLatency, getDeviceLiveToolsPing, getDeviceLiveToolsCableTest, getDeviceLiveToolsArpTable, getDeviceLiveToolsMacTable, getDeviceLiveToolsThroughputTest.

Live tools are GET-only here — you can read results, not start runs. Starting a live tool is a POST, which does not exist in this catalogue. Read existing results, or start the run from the dashboard.

Workflow: Org health check

1. getOrganizations → getOrganizationNetworks
2. getOrganizationDevices — if 0, report "no devices reported by this org" and stop. Do not present an empty inventory as a healthy one.
3. semantic_search "device availability across the organization" → invoke it
4. semantic_search "uplink loss and latency across the organization" → invoke it
5. getOrganizationLicensesOverview — licence state and expiry
6. Report health with the observation window stated. A latency number without a window is not a measurement.

Workflow: Change correlation

When something broke and nobody knows why:

1. getOrganizationConfigurationChanges — who changed what, when (one page only)
2. Correlate timestamps against the reported onset of the problem
3. getOrganizationApiRequestsOverview — was the change made by an integration, not a human?
4. Report the correlation as correlation. A config change near an outage is a lead, not a cause, and saying otherwise has sent operators down the wrong path before.

Important Rules

• No writes and no run-starting. Live tools are readable, not startable.
• 0 devices ≠ healthy. Say the org reported none.
• Capability not found on a familiar name means deprecated — search for the successor.
• Always state the observation window for any loss, latency, or availability figure.
• Record in GAIT — log every health check.

Integration with Other Skills

| Skill | How They Work Together | |-------|----------------------| | meraki-network-ops | Network and device context for health checks | | meraki-switch-ops | Port-level errors behind an uplink problem | | servicenow-change-workflow | Correlate Meraki config changes against CRs | | slack-network-alerts | Report degradation to the operator channel | | gait-session-tracking | Record all health checks |

Environment Variables

• MERAKI_DASHBOARD_API_KEY — Meraki Dashboard API key (read-only recommended)