# NLP Cloud MCP > NLP Cloud provides a high-performance API for deep text and audio analysis. It handles summarization, entity extraction, sentiment scoring, classification, speech recognition (ASR), and language translation using one MCP Server. Your agent connects to this server to process documents and media files with precision. ## Overview - **Category:** developer-tools - **Price:** Free - **Tags:** nlp, text-analysis, machine-learning, speech-to-text, summarization ## Description You gotta run your AI client through the **NLP Cloud MCP Server** when you need deep reads on text or media files. This isn't some generalized model; it’s a high-performance API that lets your agent connect directly to specialized tools for analysis, classification, and translation. You just call what you need—no complex pipeline building required. If you're dealing with massive reports, don't read the whole thing. Just hit `summarize_text` and get a short summary that keeps all the core meaning intact. It cuts those huge blocks of text down to quick, accurate reads. When your documents are messy or full of key facts, you use `extract_entities`. This tool automatically pulls out structured data—you know, names of people, dates, and company locations—straight from any chunk of writing. You get clean, usable data points instantly. Need to figure out what someone's talking about? If you feed it customer feedback, run `analyze_sentiment` and it tells you the emotional tone: positive, negative, or neutral. It’s fast and reliable for gauging public opinion. If you're dealing with audio or video files, forget manual transcription. You run `perform_asr`, and it takes that spoken word media and spits out a precise written transcript using speech recognition. It handles the heavy lifting so you don't have to listen to hours of recordings. To keep your data organized, use `classify_text`. This tool sorts text into specific categories or labels—think 'Billing Issue' or 'Feature Request.' You can even get it to categorize stuff with zero-shot learning; you don’t need to train a model for every single label. It just knows how to sort it. And when languages get in the way, you call `translate_text`. This tool handles converting written content from one language into another with high accuracy across dozens of supported languages. You're done translating and you're back to analyzing the text. Basically, your agent connects to this server, runs a specific function—say, running `classify_text` on an incoming email thread—and the server sends back structured data. It’s how you process everything from unstructured documents to media files with precision. You just point your AI client at the tool and let it work. ## Tools ### perform_asr Takes an audio or video file and generates a precise written transcript using speech recognition. ### classify_text Puts text into one of your chosen categories or labels. ### extract_entities Finds and pulls out specific types of named data, like people' names or company locations, from a block of text. ### analyze_sentiment Checks text and returns its emotional tone (positive, negative, neutral). ### summarize_text Reduces large bodies of text into short, accurate summaries. ### translate_text Changes the language of a given text block between two different languages. ## Prompt Examples **Prompt:** ``` Summarize this text using the bart-large-cnn model: [Insert Text] ``` **Response:** ``` I've summarized the text for you. Here is the condensed version: [Summary Content]. ``` **Prompt:** ``` Extract all named entities from this paragraph: [Insert Text] ``` **Response:** ``` I found the following entities: 'Apple' (Organization), 'Cupertino' (Location), and 'September 12th' (Date). ``` **Prompt:** ``` What is the sentiment of this customer review: 'The product exceeded my expectations!' ``` **Response:** ``` The sentiment analysis shows a 'Positive' result with a high confidence score. ``` ## Capabilities ### Determine text sentiment Runs `analyze_sentiment` to tell you if a piece of writing is positive, negative, or neutral. ### Categorize unstructured text Uses `classify_text` to sort text into specific, predefined labels (e.g., 'Billing Issue,' 'Feature Request'). ### Pull named entities from text Executes `extract_entities` to pull out structured data like names of people, dates, and organizations. ### Transcribe audio or video files Runs the `perform_asr` tool to convert speech from media into written text transcripts. ### Condense long documents Uses `summarize_text` to cut massive blocks of text down to a concise summary while keeping the core meaning intact. ### Translate between languages Calls `translate_text` to convert written content from one language into another with high accuracy. ## Use Cases ### Analyzing a competitor's press release A marketing manager gets a massive PDF. Instead of reading it all, they ask their agent to run `summarize_text` and then `extract_entities`. The resulting summary gives them the main claims, while the extracted entities give them specific names or product lines they can track. ### Processing multilingual customer feedback A support lead receives 50 tickets in different languages. They pipe them into an agent that first uses `perform_asr` (if audio is attached), then runs `translate_text` to get it all in English, and finally sends the result to `analyze_sentiment` to sort urgency. ### Creating a knowledge base from meeting recordings A team member uploads a 30-minute call recording. The agent runs `perform_asr` first. Then, it passes the resulting transcript to `extract_entities` to pull all key decision-makers and dates, which are then stored in a database. ### Sorting incoming news articles by topic A research analyst gets 100 daily news links. They feed the text into the agent, which runs `classify_text` to tag them (e.g., 'Finance', 'Politics') and then uses `analyze_sentiment` on each article to gauge market mood. ## Benefits - Stop reading reports cover-to-cover. Use `summarize_text` to get the core points from massive documents in seconds. - Turn speech into data instantly. The `perform_asr` tool transcribes audio or video files so your agent can actually read them and act on the text. - Structure raw, messy text with precision. `extract_entities` pulls out specific pieces of info—like dates or names—so you don't have to manually highlight them. - Understand customer mood instantly. Running `analyze_sentiment` on incoming tickets tells you if a user is frustrated before you even open the ticket. - Scale your content analysis without hiring linguists. Use `translate_text` and `classify_text` to process global, multi-lingual data streams automatically. ## How It Works The bottom line is: you call a function name, it runs the advanced model for you, and you get clean, usable data back. 1. Subscribe to the NLP Cloud server and input your API token. 2. Your AI client invokes a specific tool (e.g., `summarize_text`) and passes the required text or audio file. 3. The MCP Server executes the model, processes the data, and returns the structured output directly to your agent. ## Frequently Asked Questions **How does analyze_sentiment work with NLP Cloud?** It takes a text string and returns an emotional rating (positive, negative, neutral). It’s useful for quickly gauging customer mood from feedback or reviews. **Can I use perform_asr to transcribe video files?** Yes. The `perform_asr` tool handles both audio and video inputs. You provide the file, and it outputs a readable text transcript of all spoken content. **What is the difference between extract_entities and classify_text using NLP Cloud?** They do different things. `extract_entities` pulls out *what* specific items are (names, dates). `classify_text` tells you *what kind* of text it is (e.g., 'Support Ticket' or 'Billing Inquiry'). **How do I summarize a document using the summarize_text tool?** You pass the block of text to the `summarize_text` function. The model then condenses it, giving you the core meaning without losing context. **What are the rate limits I should know about when calling summarize_text?** The API enforces a default limit of 10 requests per minute. If your agent hits this ceiling, it will receive a 429 error code. For high-volume processing, check our batch endpoints in the documentation to manage quotas efficiently. **What file types can I pass to perform_asr for transcription?** It accepts standard audio formats like MP3, WAV, and FLAC. If you have a video file, you must first extract or encode the audio stream into one of those compatible formats before calling the tool. **How should I handle my API key when using translate_text?** You need to pass your secure API token as an environment variable or within the request headers. Never hardcode the key directly into your client script; keep it separate for security. **Does classify_text require me to train a model first?** No, you don't have to train anything upfront. You simply provide the desired predefined labels and examples in the payload. The tool handles categorization using that context immediately. **Can I summarize a long article using a specific model like BART?** Yes! Use the `summarize_text` tool and specify the `model` (e.g., 'bart-large-cnn') along with your text. You can also enable `use_gpu` for faster processing. **How do I extract names and locations from a document?** You can use the `extract_entities` tool. Provide the text and a NER model (like 'en_core_web_lg'), and the agent will return a list of identified entities such as persons, organizations, and locations. **Is it possible to transcribe audio files into text?** Absolutely. Use the `perform_asr` tool with a model like 'whisper'. You'll need to provide a JSON payload containing the audio URL or data as required by the NLP Cloud API.