How Willow Works for You
Willow sits between your AI client and your organization's tools. Every tool call your AI makes goes through Willow first. Willow checks who you are and what you're allowed to do, then routes the request to the right tool. Everything is logged.
This page explains the mental model: what the pieces are, what you control versus what your admin controls, and why things work the way they do.
The objects in Willow
MCP Servers are the tool providers: services like GitHub, Jira, or a custom internal API. Each MCP server exposes a set of actions your AI can call. Your admin decides which MCP servers are available to you.
Tools are the individual actions within an MCP server. "Create a GitHub issue" and "list open pull requests" are two different tools within a GitHub MCP server. When your AI says it has access to 47 tools, it means 47 individual actions across all your enabled MCP servers.
Skills are instructions, not actions. A skill is a Markdown document that tells the AI how to approach a task: a process, a checklist, a set of guidelines. Skills don't call APIs; they shape how the AI reasons. Skills can come from you (personal) or from your admin (org-level, automatically injected for your group). See Understanding Skills.
Commands are reusable text templates you invoke by name. Typing /code-review in your AI client expands into whatever the command contains. Commands save you from writing the same prompt repeatedly.
Toolkits bundle MCP servers, skills, and commands together for a specific workflow. When you or your admin activates a toolkit, everything in it becomes available to your AI at once.
What you control vs what your admin controls
| You control | Your admin controls |
|---|---|
| Which servers you have enabled (from what's available to your group) | Which MCP servers appear in your list |
| Your personal skills, commands, and toolkits | Org-level skills and toolkits shared to your group |
| Which individual tools within a server are active for you | Tools locked at the org level |
If a server isn't in your list, or a tool won't enable, that's an admin setting. Contact your admin.
What the dashboard shows you
The Explore section shows three counts: MCP Servers, Toolkits, and Skills available to your account.
MCP Servers: 0 means your admin hasn't assigned you to a group yet, or your group has no servers. Nothing in the UI changes this. Contact your admin.
Toolkits: 0 is normal if your organization hasn't published any.
Skills counts org-level skills assigned to your groups. If non-zero, those skills are automatically active in your AI client with no setup needed on your part.
The Discover MCP Servers section below the counts is where you browse and connect to available servers before connecting anything. Clicking Use on any card opens the Connect MCP modal pre-configured for that server.
How the connection works
MCP (Model Context Protocol) is the open standard that connects AI clients to tool servers. Willow speaks MCP on both sides: your AI client connects to Willow via MCP, and Willow connects to your organization's tools via MCP.
When you configure your AI client with a Willow URL, you're telling it to use Willow as its MCP endpoint. All the tools your org has connected flow through that single connection.
There are three ways to connect: a focused single-server connection, a dynamic multi-server connection, and a CLI option. See Connection Methods for when each makes sense.