Skip to content

DESIGN-v1.md

Latest commit

 

History

History
106 lines (70 loc) · 3.14 KB

DESIGN-v1.md

File metadata and controls

106 lines (70 loc) · 3.14 KB

ask-another v1-03

Version: v1-03 Date: 04 February 2026 Type: Specification

1. Overview

ask-another is an MCP server that supports queries to other LLMs.

It enables an MCP client (Claude, Gemini, etc.) to query other language models. Clean interface, explicit model allowlist, no provider-specific features in v1.

2. Tools

The server exposes two tools: one to discover available models and one to query them.

2.1 list_models

Returns the list of available models based on the MODEL_* configuration.

This tool takes no parameters and returns an array of model identifiers.

2.2 completion

Sends a prompt to the specified model and returns the response.

Parameter Type Required Description
model string yes Model identifier in provider/model format
prompt string yes The question or prompt to send
system string no Optional system prompt
temperature float no 0.0-2.0, defaults to 1.0

Returns the text response from the model, or an error message.

3. Model Identifiers

Models use the format provider/model-name.

Provider Examples
OpenAI openai/gpt-4o, openai/gpt-4o-mini
Google gemini/gemini-pro, gemini/gemini-1.5-pro
xAI xai/grok-beta
Anthropic anthropic/claude-3-5-sonnet-20241022
Meta (via Groq) groq/llama-3.1-70b-versatile

4. Configuration

All configuration uses environment variables in Claude Desktop config.

{
  "mcpServers": {
    "ask-another": {
      "command": "uvx",
      "args": ["--from", "git+https://github.com/matthewgjohnson/ask-another", "ask-another"],
      "env": {
        "MODEL_GPT4": "openai/gpt-4o;sk-...",
        "MODEL_GEMINI": "gemini/gemini-pro;AIza...",
        "MODEL_GROK": "xai/grok-beta;xai-..."
      }
    }
  }
}

Each MODEL_* variable defines one allowed model. Format: provider/model-name;api-key. The server scans for all environment variables matching the MODEL_* pattern on startup. Use descriptive names for the variable suffix (e.g., MODEL_GPT4, MODEL_GEMINI) rather than numbers.

5. Error Handling

The server handles errors as follows.

Condition Behaviour
Model not in allowlist Error: "Model not allowed"
Invalid MODEL_* format Error on startup, names the malformed variable
Rate limit Pass through provider's error
Invalid model Pass through error
Timeout 60 second default, return timeout error

6. Implementation Notes

The reference implementation uses Python with uv as package manager and LiteLLM as the unified LLM interface. LiteLLM provides a single completion() call that routes to 100+ providers, handling authentication and response normalisation. This avoids managing separate SDKs per provider.

7. Future Considerations

The following features are not included in v1: conversation history and multi-turn support, streaming responses, token counting and cost tracking, and provider-specific tools such as Grok search and Gemini grounding.

End of document