gpu-bridge-mcp-server

Instructions

Implementation Reference

• index.js:17-39 (registration)

  Registration of the 'gpu_run' tool, including its description and input schema.

  {
    name: "gpu_run",
    description: "Run any GPU-Bridge AI service. 30 services available: LLM inference (sub-second), image generation (FLUX, SD3.5), video generation, video enhancement (up to 4K), speech-to-text (Whisper, <1s), TTS (40+ voices), music generation, voice cloning, embeddings, document reranking (Jina), OCR, PDF/document parsing, NSFW detection, image captioning, visual Q&A, background removal, face restoration, upscaling, stickers, and more. Use gpu_catalog to see all available services.",
    inputSchema: {
      type: "object",
      properties: {
        service: {
          type: "string",
          description: "Service key. Common ones: llm-4090 (text), image-4090 (image), video (video), whisper-l4 (speech-to-text), tts-l4 (text-to-speech), embedding-l4 (embeddings), rembg-l4 (bg removal), upscale-l4 (upscale), ocr (text extraction), caption (image caption), face-restore, musicgen-l4, llava-4090 (visual Q&A), sticker, whisperx (diarized STT), bark (expressive TTS), voice-clone, photomaker, ad-inpaint, animate, image-variation, inpaint, controlnet, clip, segmentation, rerank (document reranking), nsfw-detect (content moderation), video-enhance (video upscaling), pdf-parse (document parsing)"
        },
        input: {
          type: "object",
          description: 'Service-specific input. Examples: LLM {"prompt":"...","max_tokens":512,"model":"llama-3.3-70b-versatile"}, Image {"prompt":"..."}, Whisper {"audio_url":"https://..."}, TTS {"text":"...","voice":"af_alloy"}, Embedding {"text":"..."}, OCR/Rembg/Upscale/Caption {"image_url":"https://..."}, Video {"prompt":"..."}'
        },
        priority: {
          type: "string",
          enum: ["fast", "cheap"],
          description: 'Routing priority. "fast" = lowest latency (default), "cheap" = lowest cost.'
        }
      },
      required: ["service", "input"]
    }
  },

• index.js:104-136 (handler)

  Handler logic for the 'gpu_run' tool, which makes the API call and polls for the job result.

        case "gpu_run": {
          const { service, input, priority } = args;
          const headers = {};
          if (priority) headers["X-Priority"] = priority;
          const job = await apiCall("/run", "POST", { service, input }, headers);
          if (job.error) {
            return { content: [{ type: "text", text: `Error: ${job.error}${job.hint ? `
  Hint: ${job.hint}` : ""}${job.available_services ? `
  Available: ${job.available_services.join(", ")}` : ""}` }], isError: true };
          }
          const result = await pollJob(job.job_id);
          const output = result.output;
          let text;
          if (typeof output === "string") {
            text = output;
          } else if (output?.text) {
            text = output.text;
          } else if (output?.url) {
            text = output.url;
          } else if (output?.audio_url) {
            text = output.audio_url;
          } else if (output?.embedding) {
            text = `Embedding (${output.dimensions} dimensions): [${output.embedding.slice(0, 5).map((n) => n.toFixed(4)).join(", ")}...]`;
          } else {
            text = JSON.stringify(output, null, 2);
          }
          if (result.output_notice) {
            text += `
  
  Note: ${result.output_notice}`;
          }
          return { content: [{ type: "text", text }] };
        }