generate

Instructions

Implementation Reference

• server.js:179-208 (handler)

  The async function 'generate' that executes the tool logic. It validates required 'model' and 'prompt' string arguments, constructs a request body (with optional 'system' and 'options' fields), sends a POST to '/api/generate' via httpRequest, and returns a textResult with the model's response plus timing metrics (eval_count, eval_duration_ms, tokens_per_second, etc.).

  async function generate(args) {
    const badModel = requireString(args, 'model');
    if (badModel) return errorResult(badModel);
    const badPrompt = requireString(args, 'prompt');
    if (badPrompt) return errorResult(badPrompt);
  
    const body = {
      model: args.model,
      prompt: args.prompt,
      stream: false,
    };
    if (args.system && typeof args.system === 'string') body.system = args.system;
    if (args.options && typeof args.options === 'object') body.options = args.options;
  
    const r = await httpRequest('POST', '/api/generate', body);
    if (r.error) return errorResult(r.error);
    const d = r.data || {};
    return textResult({
      model: d.model || args.model,
      response: d.response || '',
      done_reason: d.done_reason || null,
      eval_count: d.eval_count || null,
      eval_duration_ms: d.eval_duration ? Math.round(d.eval_duration / 1e6) : null,
      prompt_eval_count: d.prompt_eval_count || null,
      total_duration_ms: d.total_duration ? Math.round(d.total_duration / 1e6) : null,
      tokens_per_second: d.eval_count && d.eval_duration
        ? Math.round((d.eval_count / (d.eval_duration / 1e9)) * 100) / 100
        : null,
    });
  }

• server.js:307-326 (schema)

  Input schema and description for the 'generate' tool. Defines required properties: 'model' (string), 'prompt' (string), and optional 'system' (string) and 'options' (object for Ollama sampling parameters).

  {
    name: 'generate',
    description: 'Run a one-shot text completion against a local model (non-streaming). Returns the full response text plus timing and tokens/second.',
    annotations: { title: 'Generate text', readOnlyHint: false, destructiveHint: false, openWorldHint: true },
    inputSchema: {
      type: 'object',
      properties: {
        model: { type: 'string', description: 'Model name (e.g. "llama3.1:8b").' },
        prompt: { type: 'string', description: 'Prompt text.' },
        system: { type: 'string', description: 'Optional system prompt.' },
        options: {
          type: 'object',
          description: 'Ollama sampling/decoding options — e.g. {"temperature": 0.7, "num_predict": 100, "top_p": 0.9}.',
          additionalProperties: true,
        },
      },
      required: ['model', 'prompt'],
      additionalProperties: false,
    },
  },

• server.js:385-394 (registration)

  The HANDLERS mapping object that registers the 'generate' function (line 390) along with all other tool handlers, used by the JSON-RPC dispatch logic.

  const HANDLERS = {
    ollama_status: ollamaStatus,
    list_models: listModels,
    list_running: listRunning,
    show_model: showModel,
    generate: generate,
    chat: chat,
    pull_model: pullModel,
    delete_model: deleteModel,
  };