# Chaindesk MCP for AI Agents MCP

> Chaindesk gives you the ability to build and control custom AI knowledge agents trained exclusively on your company's private data. It lets developers programmatically manage multiple specialized bots, ingest external documents like URLs and PDFs, and query deep context-aware answers using any compatible AI client.

## Overview
- **Category:** knowledge-management
- **Price:** Free
- **Tags:** llm-training, custom-chatbots, knowledge-retrieval, no-code-ai, rag-pipeline

## Description

Building smart internal tools used to mean complicated APIs or manual data dumps. Now, you can build dedicated AI agents that act as subject matter experts for your business—all without writing complex code. This MCP lets your preferred AI client manage the whole process conversationally. You control everything from creating specialized bots configured with specific goals, to continuously feeding them fresh knowledge by adding entire websites or documents into a central datastore. Need to know what information is available? Your agent can check and monitor all your connected data sources for instant status reports. If you're connecting this through the Vinkius catalog, you get access to this full orchestration suite from one place. The result is an AI that doesn't guess; it answers using only your approved corporate knowledge.

## Tools

### list_agents
Retrieves a list of all AI bots currently configured within your system.

### list_conversations
Gets a record of past chat sessions, which can be filtered by the specific bot ID.

### list_datastores
Shows you a list of all connected knowledge collections (datastores) available to your agents.

### get_datastore
Retrieves detailed information about one specific knowledge collection by its ID.

### query_agent
Sends a message to an agent so it can answer questions using your custom knowledge base.

### update_agent
Modifies the settings or instructions of an existing AI bot.

### create_agent
Builds a brand new AI bot by providing its name, linking it to a knowledge base, and setting its core operational prompt.

### delete_agent
Removes an existing AI agent from your system entirely.

### get_agent
Retrieves all the detailed configuration and settings for a specific AI bot.

### get_messages
Fetches the complete history of messages from a particular conversation thread.

### upsert_datasource
Adds new content, like a URL or document, to an existing knowledge collection, or updates it if it already exists.

## Prompt Examples

**Prompt:** 
```
List all my current bots and check the status of our main knowledge base.
```

**Response:** 
```
**🤖 System Overview**

Here are your active agents:
*   `Support Bot`: Handles general FAQs. (Status: Online)
*   `Legal Analyst`: Only uses compliance data. (Status: Online)
*   `Product Research`: Focuses on market trends. (Status: Maintenance Mode)

**📚 Datastore Status**
The primary knowledge base (`ds_main`) is healthy and fully updated. We successfully added the Q3 reports yesterday. You're good to go.
```

**Prompt:** 
```
I need a new bot for onboarding new hires, using the HR manual data.
```

**Response:** 
```
**✅ Agent Created Successfully!**

*   **Name:** New Hire Onboarding Assistant
*   **ID:** `agent_hr123`
*   **Knowledge Source:** Linked to `ds_hr_manual`.
*   **Instructions Set:** The bot is now programmed to answer questions strictly based on HR policy and cannot hallucinate. You can start testing it right away.
```

**Prompt:** 
```
What's the latest info about our return policy? (Assume this data was added via a URL link)
```

**Response:** 
```
**🔍 Retrieval Complete**

According to the updated documentation linked from `https://company.com/returns`:
*   Returns must be requested within 30 days of purchase.
*   Items must include original packaging and proof of purchase.
*   We accept returns for a full refund, provided the item is unused.
```

## Capabilities

### Build and Manage Specialized Knowledge Bots
Create multiple distinct AI agents, assigning each one a specific role and providing core instructions to guide its behavior.

### Ingest Data from External Sources
Add or update data sources—like entire website URLs or uploaded documents—to build a real-time, comprehensive knowledge base for your agents.

### Run Contextual Queries Against Proprietary Data
Send questions to a specific agent and receive detailed answers grounded in your company’s private data, not general internet knowledge.

### Monitor AI Knowledge Bases
View the status and directory of all connected knowledge collections (datastores) directly through your AI client for quick reporting.

## Use Cases

### Handling complex customer support queries
A support agent needs to answer a question that spans three different manuals. Instead of searching three separate systems, the agent uses `query_agent` to pull context from all relevant datastores and gives one single, accurate answer.

### Onboarding new departmental knowledge
The legal team publishes a new compliance guide. The operations lead doesn't have to manually upload it; they use `upsert_datasource` on the main datastore, and all relevant agents immediately gain access.

### Debugging bot performance
A developer suspects an agent is behaving oddly. They check the conversation history using `list_conversations` and review the agent's prompt via `get_agent` to pinpoint exactly where the logic failed.

### Scaling AI services across teams
A company grows departments rapidly. Instead of building one monolithic bot, they use `create_agent` several times to build separate, dedicated assistants for HR, IT, and Sales, keeping their knowledge bases isolated.

## Benefits

- Manage your entire bot fleet from one place. Use the `list_agents` tool to see every specialized assistant you've deployed, giving you a clear view of your AI infrastructure.
- Keep knowledge current instantly. The `upsert_datasource` tool lets you add or update data sources like website URLs in real-time, ensuring your agents never use outdated facts.
- Maintain context across interactions. By accessing complete session histories via the `get_messages` tool, your agent always remembers what was discussed earlier in the conversation.
- Build specialized bots easily. Use `create_agent` to build a highly focused AI assistant for one department or topic without needing dedicated code for each one.
- Know your data sources at a glance. The `list_datastores` and `get_datastore` tools give you immediate visibility into what information is available for querying.

## How It Works

The bottom line is that you use natural conversation to manage complex AI infrastructure tasks like building bots and feeding them information.

1. Subscribe to this MCP, then grab your API Key from your Chaindesk dashboard.
2. Use your AI client to issue commands, such as asking the agent to create a new bot or add a data source URL.
3. The system processes the request, updates the knowledge base, and provides you with confirmation of the action.

## Frequently Asked Questions

**How does Chaindesk help me keep my AI bots up to date with new policies?**
You update your bot's knowledge by feeding it fresh data sources, like a URL or PDF. Instead of rebuilding the whole thing, you just use the data ingestion tools to 'upsert' the information, and the agents instantly incorporate the changes.

**Can I run multiple specialized AI bots for different teams?**
Yes. You can create separate bots, each with its own specific knowledge base and purpose. This prevents them from mixing up data or giving confusing answers across departments.

**What if my agent needs to answer questions about data that isn't in the main database?**
You must first connect the required information by using the knowledge ingestion tools. The agent can only access and report on what you programmatically feed it, ensuring accuracy.

**How do I know if my AI bots are working correctly?**
The MCP lets you monitor everything. You can list your available agents to check their status or retrieve conversation histories to review exactly how the bot responded and what data it used.

**Is Chaindesk only for developers, or can a non-technical person use it?**
While it has powerful developer features, the goal is making it accessible. You manage complex configurations through natural conversation with your AI client, meaning you don't need to write code.