# Verba MCP MCP > Verba connects your private, internal documents directly to your AI agent. It lets you treat a local knowledge base like a conversation partner; ask a question and get an answer backed up by exact document citations. You manage and query dense technical manuals or corporate policies without ever touching a web UI. ## Overview - **Category:** knowledge-management - **Price:** Free - **Tags:** rag, semantic-search, document-ingestion, knowledge-base, open-source, retrieval-augmented-generation ## Description This connector turns any isolated, text-heavy data source into something your AI agent can use in real time. Forget manually searching through SharePoint folders or exporting documents just to ask a question. With Verba, your agent reads and understands your private knowledge base—whether it's HR handbooks, technical specs, or deployment guides—and talks back using only the relevant facts. It’s like having an expert on staff who has read every document in the company, instantly. The real power comes when you connect this to other systems. If your agent pulls policy details from Verba, it can then hand those findings off to a messaging MCP to notify a team or pass them to a billing MCP for record-keeping. This kind of multi-step automation is only possible because Vinkius runs every call in its own isolated sandbox, ensuring the security and integrity of your data flow. Your agent doesn't just give you an answer; it provides citations, pointing exactly which document snippet backed up that claim. You can also manage the knowledge base itself—adding new documents or deleting outdated ones—all through chat commands, keeping your internal records current without needing a database admin to run scripts. ## Tools ### add_knowledge_document The agent ingests a new document by accepting its full text content and optional metadata JSON. ### delete_knowledge_document This permanently removes an existing document from the knowledge base; this action is irreversible. ### get_document_details The agent retrieves the complete content and all metadata associated with a specific document ID. ### get_system_config The agent pulls the current configuration settings for the entire Verba system. ### list_knowledge_documents The agent provides a list of every document ID and title indexed in the knowledge base. ### perform_rag_query This tool executes an advanced query, returning summarized answers directly from your documents along with source citations. ## Prompt Examples **Prompt:** ``` Check Verba's configuration to see which embedding model it is currently using. ``` **Response:** ``` I've retrieved your Verba system configuration. The engine is properly spun up and indicates that it is currently utilizing the 'MiniLM-v12' embedding model localized naturally. All systems appear nominal for retrieval. ``` **Prompt:** ``` Perform a RAG query asking: 'What are our key deployment steps based on the infrastructure guide?' ``` **Response:** ``` Result from Verba Knowledge Base: *"According to the 'Infrastructure Deployment v2' document, there are three key steps for deployment: 1) Initiate pre-flight health checks via CI/CD, 2) Provision ephemeral node pods for shadow tests, and 3) Cutover traffic using a blue-green weighting swap mechanism."* *(Source Citations: DocId: xfA-21v - confidence: 0.96)* ``` **Prompt:** ``` List all documents and output the unique ID of the 'Employee Code of Conduct' file. ``` **Response:** ``` I've queried the document catalog. You have 15 items overall. The specific 'Employee Code of Conduct' document is indexed under the Document ID `5fcbeb99-xxxx`. Would you like me to delete it or replace its contents with a new updated chunk? ``` ## Capabilities ### Querying private data You ask an agent a complex question and it returns a synthesized answer citing the exact source documents. ### Adding new context The agent ingests brand-new text or entire documents, making them immediately searchable by the AI. ### Tracking document inventory You list every document indexed in your knowledge base to audit what information is available. ### Inspecting specific files The agent retrieves the full text and metadata of a single, identified document. ### Checking system health You request an audit of the current Verba setup to confirm all connections and models are working. ## Use Cases ### Onboarding new hires A new employee asks the agent: 'What is the process for submitting expense reports?' The agent uses `perform_rag_query` to pull specific steps from the finance manual, citing the exact section number and preventing them from getting outdated information. ### Debugging a system failure An engineer asks: 'What are our current CI/CD standards?' The agent uses `perform_rag_query` to pull deployment protocols from the infrastructure guide, ensuring they follow the latest best practices instead of relying on memory. ### Content cleanup A Knowledge Manager runs `list_knowledge_documents`, finds an old '2018 Policy' document by ID, and then uses `delete_knowledge_document` to remove it permanently from the search results. ### Building a complex workflow The agent first performs a query using Verba. Then, because of Vinkius, it takes those retrieved policy details and passes them directly into an external messaging MCP to alert management. ## Benefits - Stop relying on keyword matching. When you use `perform_rag_query`, the agent understands intent, pulling answers from the right parts of your documents. - Keep your knowledge base clean using `delete_knowledge_document`. You can remove outdated or incorrect policies in one chat command, keeping results accurate. - Audit your entire content catalog by running `list_knowledge_documents`. This gives you a single view of every piece of indexed data. - Need to test how the system is working? Use `get_system_config` to pull live details on embedding models and cluster health, confirming everything's online. - When integrating with other tools, this MCP ensures that your agent always has access to verified source material before making a decision. ## How It Works The bottom line is that your agent uses these tools as API calls, translating natural language commands directly into structured data operations against your private documents. 1. Make sure your local or cloud Verba instance is running and you have its API URL ready. 2. Connect this MCP through Vinkius, giving the agent permission to interact with your knowledge base. 3. Tell your AI client what you need—whether it's a query ('What are our deployment steps?') or an action ('Add this new guide document'). ## Frequently Asked Questions **Can I query my local Verba instance directly through Cursor?** Yes! Once you configure `VERBA_API_URL` to point to `http://localhost:8000` (or your host port), you can prompt your AI assistant to execute rigorous `perform_rag_query` instructions without ever breaking your developer focus. **How do I insert fresh text data into Verba completely using conversational chat?** Provide the agent with your desired context directly. For example: `Add this chunk of markdown as a new document to Verba: '# Title Content...'`. The agent leverages `addDocumentTool`, serializes the payload, and commits it into Verba's vector store immutably. **Are the query answers backed by citations from its embedded documents?** Absolutely. That's the primary benefit of the integration. When you run `perform_rag_query`, Verba utilizes Weaviate's hybrid search mechanics. The output explicitly includes natural language synthesis backed by the unique document IDs and snippet texts it referenced. **What happens when I run the `delete_knowledge_document` tool?** It permanently removes data from your knowledge base. Use this carefully, as the action is irreversible; there's no undo function for deleted documents. **How do I use `get_system_config` to check my Verba setup?** It retrieves a full report on your current system configuration. This lets you confirm that the embedding model and all local LLM connections are running correctly. **What specific information can I get using `get_document_details`?** You retrieve both the full content and the associated metadata for any given document ID. This is key if you need to audit who owns a piece of data or how it was tagged. **How do I use `list_knowledge_documents` to manage my inventory?** It lists every single document currently indexed in your Verba knowledge base. This function gives you the necessary IDs and titles before you run other operations like deletion or retrieval. **What makes a query result better when I use `perform_rag_query`?** The quality relies on how dense the source material is. The agent synthesizes answers using citations, meaning every claim it makes points directly to a verified text snippet.