DeepInfra MCP by usefulapi

Most tools target clearly distinct resources and actions, but there is mild potential overlap between get_hardware and get_gpu_availability, and between get_live_metrics and get_deployment_stats. Descriptions clarify the differences well, so misselection is unlikely but not impossible.

Every tool follows the same deepinfra_ prefix and snake_case verb_noun pattern: get_, list_, start_, stop_, query_. The convention is predictable and uniform across all 18 tools.

18 tools is slightly above the typical 3-15 range, but each maps to a distinct DeepInfra REST endpoint across several domains (models, deployments, billing, usage, tokens, logs). No redundant tools are present, so the count is defensible though a bit heavy.

The surface covers reads and lifecycle control for deployments, but it lacks create_deployment and delete_deployment, and API tokens can only be listed, not created or revoked. These are notable gaps for a server meant to manage DeepInfra resources.