# Chaindesk MCP MCP > Chaindesk gives you control over custom AI agents trained on your specific company knowledge. You can build bots to answer support questions, qualify leads, and resolve FAQs using only your own documents, websites, and databases—all without writing code. ## Overview - **Category:** knowledge-management - **Price:** Free - **Tags:** llm-training, custom-chatbots, knowledge-retrieval, no-code-ai, rag-pipeline ## Description Building powerful AI requires more than just a prompt; it needs context. Chaindesk lets you connect your custom agents to deep internal data stores so they can talk about your business like an expert who's read every manual. You manage the entire knowledge base, from uploading website URLs and documents to defining how many specialized agents you need. Instead of relying on general model knowledge, these agents pull answers directly from the specific sources you feed them. Because complex automation relies on knowing what happened step-by-step, Vinkius AI Analytics provides full visibility into every single tool call and data point that flows through your agent, keeping you in control and giving you a perfect audit trail of how decisions are made. ## Tools ### create_agent Build a new, specialized AI agent by defining its name, knowledge base, and core instructions. ### delete_agent Remove an existing agent from the system entirely. ### get_agent Fetch specific details about a single AI agent, like its status or current prompt settings. ### get_datastore Retrieve detailed information on a specific knowledge collection (datastore). ### get_messages Fetch the complete text history from any recorded chat conversation. ### list_agents Get a list of every AI agent currently set up in your account. ### list_conversations Retrieve metadata for all chat conversations, optionally filtered by which agent handled them. ### list_datastores Get a list of all available knowledge collections (datastores) you've created or connected. ### query_agent Send an input message to a specific agent, triggering it to generate a response using its trained data. ### update_agent Modify the configuration of an existing AI agent's prompt or associated knowledge base. ### upsert_datasource Add entirely new data (like a website URL) or update existing content sources into your knowledge base. ## Prompt Examples **Prompt:** ``` List all my available AI agents in Chaindesk. ``` **Response:** ``` I've retrieved your AI agents. You currently have 3 active bots: 'Support Bot' (ID: agent_1), 'Legal Analyst', and 'Product Research'. Which one would you like to query or update? ``` **Prompt:** ``` Ask my 'Support Bot' (ID: 'agent_1'): 'How do I reset my password?'. ``` **Response:** ``` Querying agent... Support Bot response: 'To reset your password, go to the login page and click Forgot Password. You will receive an email with instructions.' I've documented the session ID for context. ``` **Prompt:** ``` Add 'https://vinkius.com/faq' to datastore 'ds_123'. ``` **Response:** ``` Knowledge ingestion triggered! I've added the FAQ URL to datastore ds_123. Chaindesk is now processing the content to update your agents' knowledge base. Need any other sources added? ``` ## Capabilities ### List all agents See an inventory of every custom AI agent you've built. ### Create or update agents Programmatically build, modify, and maintain your specialized AI bots with specific instructions. ### Add data sources Ingest new knowledge by adding or updating entire documents, websites, or text chunks into the system's memory. ### Query a custom agent Send a question to an agent and get context-aware answers based on your proprietary data. ### Get conversation history Retrieve the full message thread from any chat session for perfect continuity in research tasks. ## Use Cases ### Onboarding a new product line A Product Lead needs the support team to answer questions about a brand-new feature. They use `upsert_datasource` to feed all internal specs and guides, then use `create_agent` to build an 'X Feature Bot'. The agent immediately answers complex technical queries without any human intervention. ### Auditing agent performance An Operations Engineer wants to know if the 'Legal Analyst' bot is using outdated information. They use `list_datastores` and then check the data source details via `get_datastore` before running a query, ensuring accuracy. ### Fixing bad chatbot responses A Support Manager notices an agent is giving generic answers. Instead of rewriting prompts manually, they use `update_agent` to narrow the bot's system prompt and then immediately test it with a `query_agent` call. ### Building a multi-stage workflow A developer needs an agent that first checks if a user exists (`get_datastore`), then drafts a response, and finally logs the conversation history using `list_conversations` to complete the task. ## Benefits - Build specialized bots with `create_agent` that act as expert knowledge consultants, not general chat tools. They only answer questions based on the data you approve. - Stop manual content uploads; use `upsert_datasource` to programmatically feed entire websites or documents into your knowledge base in minutes. - When troubleshooting a support query, access `get_messages` to provide the agent with perfect context from previous interactions, eliminating repetitive questions. - Keep track of everything: Use `list_agents` and `list_datastores` to maintain a clear inventory of every piece of intelligence your bots use. - Need to adjust an agent's instructions? The `update_agent` tool lets you fine-tune its personality or focus without rebuilding it from scratch. ## How It Works The bottom line is that you treat your internal documentation and websites like a database, and then use AI agents as the perfect interface for searching it. 1. First, subscribe to this MCP and get your API key from your Chaindesk dashboard. 2. Next, use the agent tools to list existing agents or add new data sources by pointing them to URLs or files. 3. Finally, query an agent using your preferred client (Claude, Cursor, etc.) to let it answer questions using the knowledge you just supplied. ## Frequently Asked Questions **How do I use the query_agent tool with Chaindesk?** You send a question and an agent ID. The system then uses that specific agent's configured knowledge base to find the best answer, keeping results highly relevant and accurate. **Can I update my data sources using upsert_datasource?** Yes, `upsert_datasource` is designed to add entirely new documents or URLs while also updating existing ones. This keeps your knowledge base current with minimal effort. **What if I need multiple agents working together? (list_agents)** You use `list_agents` to manage them, and then you can chain actions: an agent might query another agent's data before generating a final answer. **Is there a way to see what the AI agent is looking at? (get_messages)** Yes. `list_conversations` helps you find a session, and then `get_messages` pulls out the full chat history so you can review exactly what was said. **What details do I need to pass when using the `create_agent` tool?** You must provide a name, a datastoreId, and a system prompt. The system prompt is critical because it defines the agent's persona and core instructions for all future interactions. **How can I check which knowledge bases are available using `list_datastores`?** The tool returns an inventory of all linked datastores, including their unique IDs and current status. This lets you confirm that the data sources required for your new agents are active and accessible. **If I need to change my agent's behavior, how do I use the `update_agent` tool?** You run `update_agent`, supplying the existing agent ID along with the specific parameters you want to modify. This allows you to refine the system prompt or name without needing to rebuild the entire assistant. **My agent seems stuck; how do I use `get_agent` to check its status?** Running `get_agent` with the ID provides a snapshot of the agent's current configuration. You can verify which datastore it is linked to and confirm that all required system prompts are intact.