# Fireworks AI MCP > Fireworks AI MCP Server connects your AI agent to high-speed generative services. Use this to perform chat completions, generate embeddings, create images from prompts, transcribe audio, and manage model lists all through one unified API. It's built for developers needing ultra-fast, reliable LLM inference and multi-modal content generation. ## Overview - **Category:** ai-frontier - **Price:** Free - **Tags:** llm-inference, generative-ai, embeddings, model-deployment, high-performance-api, ai-orchestration ## Description Fireworks AI MCP Server connects your AI agent to high-speed generative services. You can use this to chat with the server and get a conversational response using the `chat` tool. You can generate basic text continuations for a prompt or instruction using the `completion` tool. You can create multi-dimensional vector embeddings from an array of strings for semantic search using the `embed` tool. The server can generate a high-fidelity image when you give it a text prompt via the `image` tool. You can get the structural text content of an audio file by giving the `transcribe` tool a public URL. You'll also use the `list_models` tool to get a list of available model names and capabilities. ## Tools ### chat Sends chat messages to the server and gets a conversational response using Fireworks AI. ### completion Generates basic text continuations for prompts or instructions using Fireworks AI. ### embed Creates multi-dimensional vector embeddings from input strings using Fireworks AI. ### image Generates a high-fidelity image based on a text prompt using Fireworks AI. ### list_models Retrieves a list of available model names and capabilities from Fireworks AI. ### transcribe Transcribes the structural text content of an audio file provided by a public URL using Fireworks AI. ## Prompt Examples **Prompt:** ``` Chat with 'llama-v3-70b': 'Explain quantum entanglement simply.' ``` **Response:** ``` Inference complete! Llama-v3 response: 'Quantum entanglement is a phenomenon where two or more particles become connected in such a way that the state of one particle instantly influences the state of the other, regardless of the distance between them...' ``` **Prompt:** ``` Generate embeddings for these sentences: ['AI is great', 'MCP is powerful'] ``` **Response:** ``` Embeddings synthesized! I've retrieved the vector representations for your 2 sentences. You can now use these arrays for semantic search or indexing in your vector database. ``` **Prompt:** ``` Generate an image of a cybernetic forest at night ``` **Response:** ``` Image generation started! I'm using Fireworks AI inference to create your cybernetic forest visual. The high-fidelity result will be ready for you to view in just a few seconds. ``` ## Capabilities ### Generate conversations Your agent sends chat messages and receives immediate, high-speed text completions using the `chat` tool. ### Continue text prompts The agent generates basic text continuations for a prompt or instruction using the `completion` tool. ### Create vector embeddings The agent processes arrays of strings and returns multi-dimensional vector representations for semantic search using the `embed` tool. ### Produce images from text The agent sends a text prompt and receives a high-fidelity image generated by the `image` tool. ### Process audio into text The agent provides a public URL, and the `transcribe` tool returns the structural text content of the audio file. ### List and check models The agent uses the `list_models` tool to enumerate available model IDs and check model capabilities. ## Use Cases ### Building a knowledge retrieval system A data scientist needs to index 10,000 documents for RAG. Instead of writing a batch script to call a separate embedding service, the agent uses the `embed` tool, passing the document chunk array. It instantly gets the vectors needed for the vector database, keeping the entire process conversational. ### Automating content creation from media A marketing team wants to create a social media campaign. They first use the `transcribe` tool on a video meeting recording. Then, the agent uses `chat` to summarize the transcript and generate five key talking points. Finally, it uses the `image` tool to create accompanying visuals for each point. ### Debugging complex LLM prompts An AI developer is building a new feature. Instead of setting up local API keys and running manual test scripts, they use the `chat` tool to talk to the server, testing different prompts and inference parameters instantly. They can then use `list_models` to confirm the best model for production. ### Processing user-uploaded audio data A product team gets a user-submitted podcast clip. They pass the public URL to the agent, which calls the `transcribe` tool. The agent receives the clean text, which they can then immediately pass to the `embed` tool for indexing into their internal knowledge base. ## Benefits - The `chat` tool keeps your conversations running. Instead of making a separate API call for every turn, your agent manages the full chat orchestration against ultra-fast LLMs. - The `embed` tool eliminates manual vectorization. You pass an array of strings and get vector representations, ready to index for semantic search, all from a single tool call. - The `image` tool lets you skip the image API. Just give a prompt, and the agent handles the synchronous inference to deliver a high-fidelity visual asset. - The `transcribe` tool processes audio files automatically. You only need to provide a public URL, and the agent gets the clean, structural text extracted. - The `list_models` tool saves time on setup. You can query the server to list all available model IDs and check which ones are fastest for your current task. - By combining these tools, you eliminate the need to switch between multiple services. Your agent stays in one conversational flow, regardless of whether it's generating text, images, or embeddings. ## How It Works The bottom line is, your agent talks to the server, the server runs the tool, and the result gets passed back to your conversation flow. 1. Subscribe to the Fireworks AI server and input your API key into your agent client. 2. Your agent sends a request (e.g., 'Generate an image of a cyberpunk dog') and invokes the specific tool (e.g., `image`). 3. The server executes the tool, handles the inference, and returns the result (e.g., the image data or text) back to the agent for use. ## Frequently Asked Questions **How does the Fireworks AI MCP Server handle multiple model types?** The `list_models` tool lets your agent check all available models. This ensures you use the fastest or most accurate model for the job before running a task like `chat` or `completion`. **Can I transcribe audio and then embed the text using the Fireworks AI MCP Server?** Yes. Your agent calls `transcribe` with the URL, gets the text, and then immediately passes that text to the `embed` tool. It chains the process seamlessly. **Is the `chat` tool the only way to use Fireworks AI?** No. While `chat` is the primary orchestration tool, you can also call specific tools directly, like `image` or `embed`, if your agent needs to execute a function without a conversational wrapper. **What is the difference between `chat` and `completion` in Fireworks AI?** The `chat` tool manages multi-turn conversations, remembering context across multiple messages. The `completion` tool is for single, stateless text generations, like finishing a paragraph. **What kind of data does the `image` tool accept?** The `image` tool accepts a text prompt (a string). It doesn't require file uploads; the agent handles the prompt string for image generation. **How do I handle rate limits when using the `chat` tool?** The server handles rate limits using standard exponential backoff logic. If your calls exceed the allotted rate, your AI client will automatically retry the request after a calculated delay. You only need to monitor your usage dashboard. **Can I use the `list_models` tool to check which models are available for `completion`?** Yes, the `list_models` tool provides a comprehensive list of all available model IDs and versions. You can run this first to confirm the exact model name you want to use for text completion. **What data types are supported when I use the `embed` tool?** The `embed` tool accepts arrays of strings as input. It generates multi-dimensional vector representations for each string in the array. These vectors are ready for semantic search or indexing in your vector database. **Can my agent perform semantic searches using Fireworks AI embeddings?** Yes. Use the 'embed' tool. Provide a JSON array of text strings, and the agent will retrieve multi-dimensional vector representations. You can then use these vectors to perform semantic similarity matches within your database. **How do I list all available LLM and image models via chat?** Use the 'list_models' tool. Your agent will enumerate the high-speed open-source and proprietary models hosted by Fireworks AI, providing the IDs and versions needed for your inference requests. **Can I generate high-fidelity images through the agent using Fireworks AI?** Absolutely. Use the 'image' tool. Provide your text prompt, and the agent will command synchronous inference against Fireworks-hosted image models to deliver high-quality visual content natively.