Nebius Model Context Protocol Server that provides an interface to the Nebius CLI with guidance for executing Nebius service commands
The server defines 4 tools with varying quality. Tool names follow verb_noun conventions (nebius_profiles, nebius_available_services, nebius_cli_help, nebius_cli_execute), which is good. However, descriptions vary significantly in clarity and actionability. nebius_cli_execute has extensive procedural documentation but is overly verbose (700+ chars). All tools lack explicit output schema documentation in the code, return types are inferred from TypedDict hints (CommandResult, ServiceHelpResult, ServiceDescription) but not formally declared in tool definitions. Input schemas are present for some tools (nebius_cli_help, nebius_cli_execute) but nebius_profiles and nebius_available_services have empty input objects {}. Parameters that do exist (service, command) have descriptions, but lack type constraints like enums or format patterns. Error handling is present but reactive, tools catch exceptions and return error fields rather than proactively validating inputs.
Get the available Nebius services. Retrieves a list of available Nebius services including nested services.
Execute Nebius CLI command. Validates, executes, and processes the results of an Nebius CLI command, handling errors and formatting the output for better readability. You MUST NEVER execute any command that includes bash-style command substitution or piping such as `$(...)`, `| jq ...`, etc. Required Execution Pattern: 1. Use nebius_cli_help tool for getting documentation for the service before generating the command to execute. 2. Run each CLI command separately, use ONLY `nebius ...` commands. 3. Extract the necessary values from the output of the commands. 4. Use the extracted values as arguments in the next commands.
Get the Nebius CLI command documentation for the specified service. Retrieves the help documentation for the specified Nebius service. You should ALWAYS run nebius_available_services tool to list services before getting specific command documentation. You must NEVER call this tool with a service that has nested sub-services (e.g., 'iam', 'storage').
Get the available Nebius CLI profiles. Retrieves a list of available Nebius profile names from the Nebius CLI configuration file. If the user does not explicitly mention profiles, do not call this tool under any circumstances.
nebius_profiles and nebius_available_services have empty input schemas ({}). Per hard scoring rule, tools with no input schema constraints score 0 for schema. These are read-only discovery tools, but lack documented output schemas, LLMs cannot understand what structure to expect from the response.
nebius_cli_execute description is 700+ characters, exceeding the 1024-char baseline by density. While comprehensive, it contains procedural instructions that should be in server-level guidance (INSTRUCTIONS constant in config.py), not repeated in every tool invocation. This violates token efficiency and makes descriptions harder for LLMs to parse.
No output schemas formally documented. CommandResult, ServiceHelpResult, and ServiceDescription are defined as TypedDict in cli.py but not exposed in tool definitions. LLMs cannot see the structure of responses, they must infer from execution. This violates the pattern:response-shaper and forces agents to guess field names.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-22 | F | 0 | <=2025-11-25 | v2 |
| 2026-03-09 | F | 36 | - | v1 |
nebius_cli_help service parameter lacks enum or validation constraint. Description says 'e.g. applications, audit, compute, iam project, msp mlflow' but these are examples, not a formal constraint. LLMs may invent invalid service names.
nebius_cli_execute command parameter is completely unconstrained. The description forbids bash substitution and piping ($(…), | jq) but provides no regex pattern or whitelist to enforce this. Validation is runtime only, a malicious or confused LLM could still attempt injection. Input should be validated before execution with clear error messaging.
Error handling is reactive but not actionable. nebius_cli_execute catches exceptions and returns CommandResult(status='error', output='...'). However, the description does not guide the LLM on recovery, e.g., 'If --parent-id is missing, call nebius_available_services to get a valid project ID, then retry.' Agents need explicit next-step guidance per pattern:recovery-guide.