> ## Documentation Index
> Fetch the complete documentation index at: https://docs.heyhumm.ai/llms.txt
> Use this file to discover all available pages before exploring further.

# Humm MCP Server

> Connect to Humm from any MCP-compatible AI assistant and run queries without leaving your workflow.

<Note>
  The Humm MCP Server is currently in ALPHA release.  It may contain significant bugs and/or change significantly from its current implementation.
</Note>

Humm has a native MCP (Model Context Protocol) server. Connect to it from an AI assistant like Claude Desktop, Claude Code, or ChatGPT and call Humm agents directly.

<Note>
  Use MCP for querying and exploration. If you want to update memories, ontology, or commands, use the API workflows in [Context Management via API](./context-management-api).
</Note>

## What This Unlocks

* Run Humm queries from inside your AI assistant
* Automate workflows that combine Humm data with other tools
* Get answers about your customers without switching context

## Prerequisites

* An MCP-compatible AI assistant (Claude Desktop, Claude Code, ChatGPT, or any tool that supports MCP)
* A Humm account

## Setup

Humm uses OAuth2 for authentication. The exact steps depend on your AI assistant, but the flow is the same everywhere.

**Using Claude Desktop as an example:**

1. Open your MCP client settings (in Claude Desktop: Settings > MCP Servers)
2. Add a new MCP server and point it to Humm's MCP endpoint:
   ```text theme={"theme":{"light":"github-light","dark":"github-dark"}}
   https://analyst.heyhumm.ai/mcp
   ```
3. When prompted, sign in with your Humm account to authorize the connection
4. Once connected, Humm's tools appear in your assistant's available tools

## Available Tools

| Tool               | Purpose                                                                |
| ------------------ | ---------------------------------------------------------------------- |
| `list_connections` | See connected data sources / SaaaS Apps (no heavy schema)              |
| `search_tools`     | Discover executable SaaS and SQL tools, optional query filter          |
| `get_tool_detail`  | Full schema/params for a tool (`tool_name`, optional `data_source_id`) |
| `execute_tool`     | Run SQL or SaaS tools from discovery                                   |
| `search_context`   | Business context / data dictionary search (`query`, optional filters)  |

## Tips

* Pair with other MCP servers (like your code editor or project tracker) to build workflows that span multiple tools
* Use the same natural language you'd use in a Humm thread — the MCP server routes your question to the same agents

## When to Use MCP vs. API

Use the Humm MCP Server when you want an assistant to:

* Inspect connected data sources
* Discover available tools
* Search business context
* Run Humm queries without leaving your editor or assistant

Use the Humm API when you want to:

* Update memories
* Refine ontology
* Create or edit commands
* Build review-and-approve workflows around context changes