Skip to main content

Manage Tools

The Tools tab lists every tool Willow discovered from an MCP server and is where you decide what end users can actually invoke through the gateway. An enabled tool is live; a disabled one is invisible to AI clients regardless of group membership. Before users can call anything, you need to sync tools from the upstream server and enable the ones you want to expose.

Prerequisites​

You need:

Go to Build > MCP Servers. Open the three-dot menu on a server row and select Edit, or click anywhere on the row.

MCP Servers table with the row action menu open on the Everything MCP row, showing Edit, Duplicate, and Delete options

Then select the Tools tab from the server's edit page.

MCP server Tools tab showing synced tools grouped by risk category, with enabled toggles, Sync Tools, Re-assess Risk, Enable All, and Disable All actions

Sync tools​

When you first add a server, the Tools tab is empty. Sync Tools reads the current tool list from the upstream MCP server and populates the tab. Run it again whenever the upstream server adds, renames, or removes tools.

Select Sync Tools when:

  • the table is empty after adding a server
  • the upstream MCP server added or renamed tools
  • a tool call fails with "tool not found" in logs

After syncing, confirm the expected tools appear before enabling them for users.

If the upstream server renames a tool, the new name appears as a new tool on the next sync, and the old name is marked Deleted or removed. Calls to the old name fail, so make sure the new tool is enabled and turn off the old one. With Auto Sync Tools on you do not have to run the sync yourself; see Proxy Configuration.

Long tool names​

Willow exposes each tool to MCP clients as <server slug>__<tool name>, and many clients reject long names. When the server slug and the tool name together reach 60 characters, Willow serves the tool under a shortened name that ends in a short hash, and keeps the upstream name so calls still reach the right tool. The row shows a Shortened badge with both names. Shortening happens on Sync Tools, and the sync re-enables a tool that was previously disabled only because its name was too long. Until you sync, an enabled tool whose stored name is too long is hidden from MCP clients, and the Tools tab warns you.

Enable and disable tools​

Each tool row has an enable toggle. Disabled tools are not exposed to AI clients through the gateway.

Tools are grouped by risk category into Read-only tools and Write/delete tools. Each group is collapsible, carries its own enabled-of-total count (for example Read-only tools (19/20)), and has its own Enable All and Disable All controls, so you can selectively expose read-only tools while keeping write-heavy ones off until you're confident in access controls, logging, and guard behavior.

Granting different groups different tool subsets

Tool access is configured per MCP server instance, not per group. To give one group a different subset of tools than another from the same server, add the server a second time — use Duplicate from its row menu to copy the configuration — enable only the intended tools on each instance, and assign each instance to its target group. Members of both groups get the union of what both instances expose.

Tool states​

As you sync, edit, or remove tools, row badges track each tool's relationship to the upstream definition.

BadgeMeaning
AvailableThe upstream server offers the tool but it isn't added to Willow yet.
ModifiedThe tool saved in Willow differs from what the upstream server offers now. Either you edited it in Willow, or the server changed it since the last sync. Hover over the badge to see what differs: name, description, arguments, action method or mapping.
DeletedThe tool no longer exists upstream but remains listed until removed or reset.

A tool with no badge matches the upstream definition.

Use Reset to default from the row menu to replace a modified tool with the upstream server's current definition. It changes only that tool and keeps whether it is enabled and whether it requires approval. To pick up one changed tool without touching the rest, use it instead of Sync Tools, which re-reads every tool and asks you to confirm because it can replace your edits.

Require approval for a tool​

Some tools are too consequential to run without a human review. Marking a tool as requiring approval adds a confirmation step before the AI client can invoke it.

  1. Open the row menu for a tool.
  2. Select Require approval.

The row shows an Approval Required badge. The menu action changes to Remove approval requirement when approval is already enabled.

When an AI client calls the tool, the call is held and the client gets a link to an approval page, where the user approves or declines. AI tools that support MCP Apps, such as Claude, can show Approve and Decline buttons in the chat instead; see Approvals Inside the Chat. Tools that don't support MCP Apps always get the link. After the user approves, the AI client calls the tool again and it runs. An approval covers the same user, the same tool, and the same arguments for a limited time, so repeating that exact call doesn't prompt again, but a call with different arguments does. Admins set how long an approval stays valid under Approval Validity Period.

Approval applies to every call to the tool. When only some invocations are consequential — a specific Slack channel, a labeled Drive file, a restricted Jira project — gate those instead with Conditions.

Test a tool​

Before rolling out a new MCP server to users, run a test call to confirm end-to-end connectivity: from Willow's gateway through authentication to the upstream server and back.

  1. Open the row menu for a low-risk tool.
  2. Select Test Tool.
  3. Enter the smallest valid arguments in the JSON editor.
  4. Select Run test.
  5. Read the result panel for structured output or an error.
Test Tool modal with arguments JSON editor on the left, result panel showing Ready to run on the right, Reset sample action, and Run test button

After running, the result panel shows the raw JSON response from the upstream server.

Test Tool modal after a successful run, showing JSON response in the result panel

If the test fails, the error message in the result panel usually points directly at the cause:

Error typeCheck
Missing or invalid argumentsAdjust the test input to match the tool schema.
Authentication errorReturn to Configure MCP Server Authentication.
Upstream connection errorCheck MCP configuration, base URL, transport, and network reachability.
Missing toolRun Sync Tools again.

Edit a tool​

Willow lets you override a tool's name, description, schema, and mapping without touching the upstream server. This is useful for tightening an overly permissive schema, renaming a tool to match your internal naming conventions, or adding an annotation that affects guard behavior.

Open the row menu and select Edit to open the tool's full edit page.

Tool edit page showing Name, Description, Annotation collapsible, Schema step with Input Schema JSON editor, Mapping step, Form/JSON toggle, and Save button at top with back arrow

The editor has two steps. In Schema (step 1), edit the Input Schema JSON that defines what arguments the tool accepts. Select Next to advance to Mapping (step 2), where you configure how Willow transforms arguments before forwarding the call upstream. Switch between Form and JSON mode at any time using the toggle at the top. Expand Annotation to adjust tool annotation settings. Use ← Back to return to the Tools tab without saving, or Save to commit your changes.

Editing a tool marks it as Modified in the tool list. For field-level detail, see MCP Server Settings Reference.

Re-assess risk​

Use Re-assess Risk on the Tools tab or from an individual tool's row menu to refresh risk scoring. Risk bands affect which tool group a tool appears in and can influence guard rules. If no visible change appears after running re-assess, the action may still have completed in the background. Check the tool's group assignment in the table.