# Deep Talk MCP > Deep Talk analyzes massive streams of conversation data from sources like Zendesk or Intercom. It lets your agent automatically pull key customer topics, measure sentiment (positive/negative), and group similar support interactions into actionable clusters. Stop sifting through thousands of transcripts; Deep Talk delivers structured insights instantly. ## Overview - **Category:** customer-support - **Price:** Free - **Tags:** sentiment-analysis, conversation-mining, topic-modeling, customer-insights, nlp, data-clustering ## Description Processing large-scale customer conversations used to take dedicated data science teams days. Now, you can connect the Deep Talk MCP directly to your AI workflow. It reads everything—from chat logs to support tickets—and automatically pulls out what matters: the specific topics customers are complaining about, how positive or negative their tone is, and groups of similar feedback that reveal patterns. This means product managers get instant theme reports, and CX teams pinpoint common pain points without running complex SQL queries. Vinkius hosts this MCP so your agent can access these insights right where you're working. You just tell your AI client to analyze a dataset, and it handles the rest. ## Tools ### create_batch_prediction Requires the dataset ID, pipeline name, and the column containing the text to analyze. Start a batch prediction job on a dataset ### get_batch_prediction Get the status and results of a specific batch prediction ### get_pipeline_details Get configuration details for a specific pipeline ### list_batch_predictions List all batch prediction jobs and their statuses ### list_pipelines List all NLP analysis pipelines in your Deep Talk account ## Prompt Examples **Prompt:** ``` List all conversation datasets currently processed. ``` **Response:** ``` I've found 5 datasets in your account, including 'Zendesk Support Q2' and 'Intercom Feedback June'. Both are 'Completed'. Would you like to see the extracted topics for 'Zendesk Support Q2'? ``` **Prompt:** ``` Show me the top topics identified in the 'Customer Feedback' dataset. ``` **Response:** ``` In the 'Customer Feedback' dataset, the top 3 topics are: 'Pricing Plans' (45%), 'Feature Requests - Mobile App' (30%), and 'Login Issues' (25%). Should I list the sub-topics for any of these? ``` **Prompt:** ``` What is the sentiment summary for our recent support interactions? ``` **Response:** ``` The sentiment analysis for your latest support dataset shows 65% Positive, 20% Neutral, and 15% Negative interactions. Sentiment has improved by 10% compared to last month. Would you like to see the clusters with the most negative sentiment? ``` ## Capabilities ### Monitor account usage Check your Deep Talk account status, including remaining processing credits and user permissions. ### List available datasets See a list of all conversation data you've uploaded, along with their current analysis progress. ### Get dataset status details Retrieve specific metadata for one dataset, confirming its source and whether NLP clustering is complete. ### Analyze sentiment scores Generate a summary showing the proportion of positive, neutral, and negative tones across your entire data set. ### List connected sources Review all external platforms (like Intercom or Zendesk) linked to Deep Talk and how much data is flowing from each. ### Identify conversation groups Pull lists of conversations that are semantically similar, grouping them into distinct clusters with key associated keywords. ## Use Cases ### The Product Team Needs a Roadmap Priority A PM asks the agent to 'Show me the top topics identified in our support chats.' The agent runs `list_extracted_topics`, immediately providing data that shows 'Pricing Plans' is 45% of all complaints, instantly proving where development focus should go. ### The CX Lead Needs a Sentiment Health Check A CX lead needs to know if the recent update caused issues. They ask for `get_sentiment_analytics` on the last 30 days of data, and the agent reports a sharp drop in positive sentiment, allowing them to issue an immediate fix. ### The Data Analyst Needs Contextual Validation An analyst wants to verify if 'Login Issues' are a persistent problem. They use `search_topics_by_keyword` with the term 'login', and the agent returns all related themes and their frequency, confirming it's a systemic issue. ### The Operations Manager Needs Source Audit An ops manager suspects data gaps. They run `list_connected_sources` to verify that both Zendesk and Intercom are properly connected and syncing the expected volume of records, ensuring no blind spots. ## Benefits - Pinpoint pain points fast. Instead of reading through thousands of tickets, use `list_extracted_topics` to get a definitive list of the top themes driving customer frustration. - Track emotional trends over time. Use `get_sentiment_analytics` to measure if your product changes are actually improving user satisfaction, quantifying the shift from negative to positive tones. - Group similar feedback automatically. The `list_conversation_clusters` tool identifies conversations that mean the same thing but were written differently, helping you find hidden patterns in customer complaints. - Keep track of data sources. Run `list_connected_sources` to confirm which platforms are feeding data and if synchronization is running smoothly across all your channels. - Know exactly what's happening. Check the status of large jobs using `list_processing_tasks` before relying on complex analyses, ensuring your results aren't based on incomplete data. ## How It Works The bottom line is you get actionable insights from raw text without writing any complex pipelines. 1. Connect the Deep Talk MCP to your AI client and authorize it using your API Key. 2. Tell your agent what you want to analyze, for example, 'List all conversation datasets currently processed' or 'Get a sentiment summary'. 3. Your agent executes the tool, and the response returns structured data—like topic lists or percentage breakdowns—ready for immediate use. ## Frequently Asked Questions **How do I check if my conversations are ready for analysis using list_analysis_datasets?** You use `list_analysis_datasets` to see all uploaded data, and then you check the status column. If it says 'Processing' or 'Pending,' wait until the job is complete before analyzing. **What tool should I use to find out if negative sentiment has increased?** Use `get_sentiment_analytics`. This tool provides a summary of positive, neutral, and negative scores across your entire data set, letting you compare periods easily. **Can Deep Talk help me find specific complaints using search_topics_by_keyword?** Yes. You run `search_topics_by_keyword` with the term you need (like 'checkout'). The tool returns only themes matching that keyword, filtering out noise. **How do I know which NLP models are available for my data?** Run `list_available_nlp_models`. This shows every model your agent can use to categorize conversations, including sentiment and intent tools. **How do I check my remaining processing credits or user roles using get_account_details?** It provides immediate metadata on your account's current status. You can use get_account_details to view subscription tiers, see how many processing credits you have left, and confirm the various user roles set up in your Deep Talk account. **What does running list_connected_sources show about my data integrations?** It gives a clear overview of every external platform connected to your account. The tool lists these sources, confirms their synchronization status, and tells you the total volume of data ingested from each one. **If an analysis job fails or stalls, how do I monitor its progress using list_processing_tasks?** The tool gives you a real-time list of all active processing jobs. You can check the status of both ingestion and NLP analysis tasks, seeing their current completion percentages to determine if they are stuck. **When I use list_conversation_clusters, what information does it provide about similar conversations?** It returns groups of semantically similar conversations identified through unsupervised learning. This data includes the size of each cluster and representative keywords that define that group's topic. **How do I get a Deep Talk API Key?** Log in to your Deep Talk account, navigate to the API section in your settings, and you can generate or retrieve your unique API Key from there. **Can the agent process real-time conversations?** This integration currently focuses on analyzing datasets that have already been uploaded and processed within Deep Talk. Real-time streaming analysis is managed via the Deep Talk dashboard or webhook integrations. **What languages are supported for analysis?** Deep Talk supports multiple languages for NLP analysis, including English, Spanish, Portuguese, and French. The agent retrieves results based on the analysis performed in your account.