How to Use AI and Automation in Pumble: API, AI Assistant, and MCP Server

Want to automate workflows, build custom messaging alerts, and connect LLMs directly to your team communication data? You can do so using Pumble by CAKE.com’s 3 core engineering interfaces:

1. The Pumble API addon — generate an API key, send messages, react, and create channels via HTTP requests.
2. The Pumble AI Assistant addon — bring an AI provider into the workspace using your own API key.
3. The Pumble MCP server — connect external AI agents (Claude, Gemini, Codex) to Pumble so they can search, read, and act.

These developer surfaces allow technical teams to integrate custom code or external AI agents with Pumble while completely avoiding per-seat AI premiums. 

Here’s how you can use Pumble’s AI infrastructure to automate processes and create a context-aware collaborative space.

Setting up the Pumble API addon

The Pumble API addon provides simple HTTP endpoints to handle messaging actions inside your workspace. It is available on all Pumble plans to Owners, Admins, and Members.

However, it doesn’t support an interactive message builder framework or an Events API. 

It is built to effectively handle:

• Generating unique workspace API keys 
• Sending messages
• Deleting messages 
• Adding reactions or emojis in active threads 
• Creating new channels via standard POST requests

This focused scope keeps the API straightforward to implement and maintain.

Authentication and scopes

To interact with the API, you must first generate an API key within your Pumble workspace administration panel. This token must be passed inside the HTTP header of every request as a bearer token.

You first need to install the API addon on your Pumble workspace:

1. In the left sidebar, click + Add apps. 
2. Select Install in the API section.
3. Allow permissions.

Now, you can generate API keys directly in Pumble: 

1. In the left sidebar, click API. 
2. Select Add API key.
3. Enter API key name.
4. Click Generate to complete the action.  

Keep in mind that the rate limit is up to 1000 requests per minute per user. 

You can find the complete list of available actions in the Swagger documentation. To access the documentation in Swagger, enter your API key in the Authorize part. 

Moreover, you can use the following commands directly in messages to interact with the Pumble API: 

• /api-keys generate — to generate API keys for particular users and store them in their database. You can generate as many API keys as you want to.
• /api-keys list — to list all users’ API keys.
• /api-keys help — to list all valid commands you can use.

Posting build notifications

To integrate Pumble into your continuous deployment pipeline, you can dispatch automated build alerts using a standard curl command. 

The structure requires a destination channel ID and a text string like in the example below: 

This returns a JSON response containing the message metadata, which you can parse to track message states or append subsequent threaded replies.

Automate your workflows in Pumble

Using the Pumble AI Assistant addon

The Pumble AI Assistant addon embeds an LLM interface directly into your team chat interface, letting you:

• Retrieve knowledge instantly.
• Automate routine admin tasks.
• Maintain context across multiple tools.

This Pumble functionality doesn’t cost any hidden per-user AI markup fees on top of your primary chat subscription. Instead, Pumble uses a bring-your-own-key (BYOK) pricing model, which means the AI infrastructure provider invoices you directly for the raw tokens your team uses. By using a BYOK model, you control your processing infrastructure, budget, and data access policies directly.

Installation and provider setup

Here’s how to install your Pumble AI Assistant addon in 4 steps: 

1. Navigate to your workspace administration panel. 
2. Select Configure provider under the AI Assistant section. 
3. Choose your preferred foundational model provider. 
4. Paste your API key into the secure credential field.

To learn how how to make the most of this Pumble feature, check out the video below:

Provider support 

For the AI Assistant addon to work, you need to have an active account with a supported LLM infrastructure company. 

The live provider support is currently available for Google Gemini and Open AI ChatGPT. Anthropic Claude, Perplexity, and Grok are in the future roadmap. 

It’s important to note that Claude doesn’t function as an in-channel AI assistant provider yet. Claude is currently handled exclusively through the separate MCP server layer. 

In-channel reactions

The Pumble AI Assistant works in 2 different operational models inside the Pumble workspace: 

1. Private DMs. By sending a direct message to the Assistant, you initiate a single-user private chat session. The replies are confidential and visible only to you. 
2. Channel mentions. When you tag the Assistant in a public or private channel using its @ handle, it then responds inline by creating a structured thread. The reply remains fully visible to all members who have access to that channel. 

To protect intellectual property and eliminate data leaks, the Pumble AI Assistant operates under a strict scoped-context guarantee. It doesn’t passively read background channel traffic. Unless explicitly tagged using a @mention, it can’t read any messages.

Collaborate securely in Pumble

Thread summarization

To break down long discussions quickly without reading every message, you can use the built-in slash command /ai-summarize. 

The system analyzes the entire conversation history of that specific thread and returns an organized summary. 

As for the security rules for this command, they will adapt to the channel type: 

• Public channels don’t require you to manually add or invite the Assistant user beforehand, as the slash command works automatically. 
• Private channels and group DMs need to have the AI Assistant user explicitly added as a member of that private group conversation before you can use the command. 

Reduce context switching

Stay focused on your tasks while your Pumble AI Assistant handles the gruntwork.

Sign up Learn more

Connecting AI agents with the Pumble MCP server

Pumble maintains a native MCP (Model Context Protocol) server hosted directly at pumble.com, helping you connect your workspace to your AI agent of choice.  

While third-party automation tools like Zapier can handle high-level application routing, the native MCP server provides direct, low-latency access to workspace operations. This allows external engineering agents, such as the Claude Command Line Interface (CLI), to: 

• Run deep message history searches. 
• Read content states. 
• Execute administrative functions on your behalf.

Now that we’ve covered how the Pumble MCP server works, let’s break down its actual setup process. 

Workspace credential configuration

Before setting up a local agent configuration file, you must build a custom integration app inside your workspace panel to generate the necessary keys.

To do so, follow this simple 5-step process: 

1. Open Workspace settings and select Configure apps.
2. Click Create app and choose the Create from scratch option.
3. Assign a distinct app name and a functional description. 
4. Define your required User Scopes and Bot Scopes, depending on the automation tools your agent needs to execute. 
5. Click Install to deploy the integration to your workspace.   

Once installed, copy the 3 primary credentials available in the app dashboard:

• The App key
• The User token
• The Bot token

Client integration configurations

To connect an external command-line agent to the native Pumble MCP server endpoint, you have to add the appropriate configuration blocks to your local developer tooling files. Note that the custom header x-app-token must match your generated App key.

Now, let’s walk you through connecting the Pumble MCP to different available AI agents.  

Claude configuration

To connect Claude CLI to the Pumble MCP server, paste the code below into your .mcp.json file: 

Or, you can use the following command to install automatically via the terminal:

claude mcp add pumble-mcp https://mcp.pumble.com/mcp –transport http –header “token: token” –header “x-app-token: xpat-token”

Note: The x-app-token/xpat token is the App key you receive when you create an app from scratch. 

Codex configuration 

For Codex engineering agent layers, paste the code below into your config.toml file: 

Gemini configuration

To connect your Pumble workspace context to Google Gemini, paste the code below into your settings.json file: 

Alternatively, you can use the following command: 

Native capabilities and available tools  

When you connect an external LLM agent through the Pumble MCP server, it queries the available tools. The specific actions the agent can perform depend entirely on the checkbox scopes you selected during the app registration phase. 

Available tool list 

Here’s a list of all available tools your agent has access to on the Pumble MCP server, complete with their definitions.  

Moreover, you can use the following native MCP server built-in prompts to complete complex actions quickly. 

Which one should you choose?

To help you make the most informed decision, here’s a reference table to map your requirements to the most suitable Pumble automation interface. 

FAQ

Can the Pumble AI Assistant summarize chat threads?

Yes. Enter the /ai-summarize slash command inside any conversation thread. The Assistant will break down the past text and output an organized summary. This works automatically in public channels. For private channels and group DMs, the Assistant must first be added as a member of that specific channel.

Can the Pumble AI Assistant read all my messages? 

No. The AI Assistant operates under a strict scoped-context framework. It cannot read channel history passively. It can only load and process text from messages where it is explicitly mentioned using a @mention.

Does Pumble support Anthropic Claude as an in-channel assistant? 

No. Pumble’s in-channel AI Assistant addon currently supports OpenAI ChatGPT and Google Gemini. Support for Anthropic Claude, Perplexity, and Grok is planned for future updates. To use Claude with Pumble today, connect via the native Pumble MCP server using your local Claude command-line toolchain. 

Create smarter workflows in Pumble by CAKE.com

Integrating AI Assistants, custom APIs, and the MCP server transforms Pumble from a team chat app into a powerful developer hub that reduces context switching and automated manual workflows. 

This way, Pumble lets you streamline complex backend tasks and allows your engineering team to build a faster and more context-aware environment inside the chat interface.

Explore more about Pumble’s integration capabilities! 

Try Pumble today