Integrations

How integrations work

Waffle AI connects to external tools through MCP, the Model Context Protocol. An admin connects a tool, scopes it to the channels that should use it, and chooses read only or read and write. Tokens are encrypted at rest, so an integration is a scoped colleague, not an open door.

Waffle runs the MCP client on its own infrastructure, so your tokens never leave Waffle and the AI can use any tool that speaks the protocol. Connecting a tool tests it before it is saved, and each connector shows up to Waffle AI only in the channels you allow. The same conversation that hosts your team now reaches your code, your issues, and your docs. That matters because tools have multiplied: the average company used about 106 different SaaS applications in 2024 (BetterCloud, 2024), and jumping between them is not free. One study of Fortune 500 teams found workers toggle between applications roughly 1,200 times a day and lose close to four hours a week, about 9 percent of their time, just reorienting after each switch (Harvard Business Review, 2022). Reaching those tools from inside the chat you already have removes some of those switches.

Waffle AI also reads and writes files directly. Share a PDF, spreadsheet, or wall of text and it answers with specifics, and it can write and attach real Word, PDF, PowerPoint, Excel, CSV, and Markdown files without any connector at all.

Admin control

Admins decide exactly what Waffle AI can reach. Connectors are scoped per channel, each connector is set read only or read and write, tokens are encrypted at rest, and Waffle AI can be turned on or off per channel. Writing is off by default, so nothing changes until you allow it.

• Per channel scoping: a channel uses all connectors, none, or a hand picked set.
• Read only or read and write, chosen per connector, off by default.
• Connector tokens encrypted at rest with AES-256-GCM, never shown again after setup.
• Waffle AI turned on or off per channel, with a per channel persona.

Frequently asked questions

What integrations does Waffle support?

Waffle AI connects to GitHub, Linear, and Atlassian (Jira and Confluence), plus any custom MCP server, through the Model Context Protocol. It also reads the files you share (PDF, spreadsheets, long documents) and writes real files: PDF, Word, PowerPoint, Excel, CSV, and text or Markdown. An admin connects each tool and scopes it to the channels that should use it.

What is MCP?

MCP is the Model Context Protocol, an open standard for connecting AI assistants to external tools and data. Waffle runs a server side MCP client, so any tool that speaks MCP can be connected. Your tokens stay on Waffle infrastructure, encrypted at rest, and are never shown again after setup.

Can I connect my own internal tool?

Yes. Alongside the GitHub, Linear, and Atlassian presets, an admin can add any custom MCP server by URL. Waffle tests the connection before saving it, blocks private and internal network addresses, and encrypts the token at rest. Once connected, its tools are available to Waffle AI in the channels you scope it to.

Can I control which channels use an integration?

Yes. Every connector is scoped per channel. A channel can use all connectors, none, or a hand picked set, so a sensitive channel can stay closed while an engineering channel reaches GitHub and Linear. Admins set this when editing a channel, and it only ever removes tools, never adds them.

Are my integration tokens safe?

Connector tokens are encrypted at rest with AES-256-GCM and are never shown again after setup, nor returned by any API or written to logs. Anything a connected tool returns is treated as untrusted input and never as instructions. Writing is off by default and enabled per connector, and access is scoped per channel.

Sources

Statistics on this page are drawn from the following public sources.

• BetterCloud, SaaS statistics (average SaaS applications per company), 2024
• Harvard Business Review, How Much Time and Energy Do We Waste Toggling Between Applications, 2022