Cisco MCP Servers Bring Network Data to AI Agents
2 min readCisco has released Model Context Protocol servers for Meraki and Catalyst Center, giving AI agents a supported path to live network data. The Cisco MCP servers are in beta and read-only, and two of the three are already published as open source on GitHub.
What MCP Changes for Network Data
Model Context Protocol is an open standard for connecting AI models to outside tools and data sources. Without it, feeding network state into an assistant meant writing custom API glue for every platform, then maintaining that glue as the APIs shifted underneath it.
Cisco already ships an AI Assistant inside Meraki for teams troubleshooting within that product. MCP points the other direction. It pulls network context out into whatever workflow an organization is building, whether that is a ServiceNow queue, an employee self-service portal, or a cross-vendor operations agent.
Three Deployment Paths
Writing on the Cisco developer blog, product marketing manager Alex Kim and engineering product manager Oren Brigg laid out three options. A Cisco-hosted Meraki server absorbs the operational overhead for teams that want the fastest start. An open-source Meraki server lets organizations run and modify the code themselves. An open-source Catalyst Center server deploys locally, exposing inventory, device health, wireless experience, software, and compliance data.
The rule of thumb the authors offer is simple. Use AI Assistant when the work starts inside the Cisco product, and use MCP when the work starts in the workflow built around it.
Why Cisco MCP Servers Matter
The example Cisco leads with is the vague ticket. Someone reports that the Wi-Fi is slow, and a first-line analyst spends the next stretch of time establishing whether the problem is one device, one access point, or one building. An agent with read access to that data can attach the evidence before the ticket is ever escalated.
The local deployment story may matter more. Regulated industries, government networks, and air-gapped environments cannot send operational data to a vendor-hosted AI service. Running the Catalyst Center server inside the same boundary as the controller is a workable answer. Cisco is careful to note the limit: self-hosting the Meraki server does not make the Meraki Dashboard air-gapped, because live data still requires a connection to the cloud.
Read-only and beta are real constraints, and nothing here pushes a configuration change. The thing to watch is whether Cisco extends these servers past reads.