Fiber

Fiber is a local MCP server for optical network operations in AI data centers, giving an assistant read and optional write access to a FIBER deployment's teleop

What you can do

Read operational data from a FIBER deployment: health, cluster overview, alerts and other read tools (24 in total for a viewer role). Inspect optical and network telemetry such as gNMI data, ML-based remaining-useful-life predictions, Digital Twin simulations and NCCL ingest status. Query semantic data through SPARQL. When write tools are enabled, trigger a demo, run NCCL ingest and create an RMA. Scope which tools are exposed with an allowlist.

When to use it

Useful when you operate or monitor optical infrastructure in AI data centers and want an assistant to pull health, alert, telemetry and predictive-maintenance information from a FIBER deployment, or to run limited write actions such as RMA creation.

Requirements

Runs locally as a Python package (fiber-mcp) over stdio on a desktop. Needs a FIBER API key in FIBER_API_KEY, whose role (viewer, operator, admin) determines which tools work. Optionally set FIBER_API_URL for the deployment base URL, FIBER_ENABLE_WRITE=1 to expose write tools, FIBER_HTTP_TIMEOUT, and FIBER_MCP_TOOLS_ALLOWLIST. Network access to the FIBER API is required.