Skip to main content

Willow Admin

Willow Admin exposes Willow's own Admin API as an MCP server. Adding it lets an AI client read and change your Willow organization the same way an admin would in the dashboard: users, groups, MCP servers, commands, skills, toolkits, guards, policies, background agents, audit logs, and custom analytics.

This is the one connector whose upstream service is Willow itself. Treat it accordingly: a caller with the write tools enabled can reconfigure the gateway that governs every other integration.

Add it​

  1. Open Build > MCP Servers > Add MCP Server.
  2. Search for Willow Admin and hover the card.
  3. Select Use.
  4. Review the risk level and the tool list, then select Create.

The connector is scored Medium Risk by default. At that level Willow enables 52 of its 60 tools and holds back the eight delete tools. Raising the risk level in the add modal includes them; see Manage tools to change the selection later.

Available tools​

The connector exposes 66 tools. They divide cleanly by what they do to your organization.

Read-only​

List Users, List Groups, List MCP Servers, List Commands, List Skills, List Toolkits, List Guards, Query Logs, Query Admin Logs, Get Custom Analytics Guide, Get Custom Analytics Catalog, Generate Custom Analytics Query, Run Custom Analytics, List Custom Dashboards, Get Custom Dashboard, List Discovered MCP Servers, List Discovered Skills, List Discovered Agent Memory, List Discovered Plugins, List Devices, Get Discovered Device, List AI Agents, List Discovered CLI Tools, List Discovered Web AI Agents, List Discovered Vibe Apps, List Discovered Auth Connections, Get Discovery Statistics, Get Enforcement Policy, List Background Agents, Get Background Agent, Get Background Agent by Slug, List Background Agent Sessions, Get Background Agent Session Events

List MCP Servers takes an optional search argument that filters servers by name or slug, and an include_tools argument that adds each server's enabled tools (name, slug, description, and risk). Combine them to keep the response small.

Query Logs reads the audit log and the agent telemetry. Its action argument takes init, connect, tool_call, connector_auth, webhook, guard_check, add_skill, agent_tool_call, or agent_event, or a comma-separated list such as agent_tool_call,agent_event. Besides the user, client, MCP server, tool, and guard filters, it filters agent telemetry by event_name, skill_name, otel_source, device_id, session_id, and is_background_agent, and tool calls by toolkit_slug, tool_call_tool_prefix, and tool_call_error_code. Without fields, every column comes back, including the large data payload. Pass fields with a comma-separated list of columns to keep the response small; id and created_at are always returned. limit is at most 1000. To read further pages, pass before_created_at and before_id from the previous response's pagination.next_cursor. A cursor cannot be combined with skip. For counts and breakdowns, use Run Custom Analytics instead.

The four custom analytics tools run the same engine as Explore. An AI client reads the guide, then either generates a Willow Query Language (WQL) query from a plain-English question or writes WQL itself, and runs it. Run Custom Analytics returns JSON by default and CSV when asked. Get Custom Analytics Catalog and Generate Custom Analytics Query return an error when Explore is not enabled for your organization, and Generate Custom Analytics Query also needs an organization or platform LLM key.

The four custom dashboard tools manage the dashboards under Dashboards. List Custom Dashboards and Get Custom Dashboard read them. Create Custom Dashboard and Update Custom Dashboard take a title and visuals, where each visual has a title plus a WQL query or a spec from Generate Custom Analytics Query, and need the admin:write scope. Both return the dashboard and an absolute url that opens it in the admin UI. An update changes only the fields you pass, and passing a layout replaces the current visuals.

Create and update​

Create User, Update User, Create Group, Update Group, Add Integrations to Group, Add Toolkits to Group, Create MCP Server, Update MCP Server, Create Command, Update Command, Create Skill, Update Skill, Create Toolkit, Update Toolkit, Promote Toolkit, Create Custom Dashboard, Update Custom Dashboard, Create Guard, Update Guard, Update Enforcement Policy, Create Background Agent, Update Background Agent, Sync Background Agent, Invoke Background Agent, Rotate Background Agent Secret

Create Skill and Update Skill run the same security review as the dashboard. New skills are saved as drafts unless the call asks to publish them. The response carries a review.status of published, draft, pending (held as a hidden draft until an admin approves it), or rejected (blocked by a guard policy), and the tool descriptions tell the AI client to report a pending or rejected skill to the user. Changing a skill's content, scripts, or references, or publishing it, runs the review again; a metadata-only edit does not. See Review states.

Delete​

Delete User, Delete Group, Delete MCP Server, Delete Command, Delete Skill, Delete Toolkit, Delete Guard, Delete Background Agent

These eight are classified high risk and are excluded at the default Medium Risk level.

Who can call it​

Organization owners can call every tool. Any other admin can call the tools their role allows: the role needs the same permission as the matching dashboard page, so a custom role with Audit > View Logs can run Query Logs. A call without the permission returns a 403 that names the missing one. API tokens are not limited by roles; they use their own scopes.

Publishing is a separate permission. Setting a skill or toolkit to active needs Publish on Skills or Toolkits, and an MCP server created through the connector by a role without Publish on Integrations (MCPs) is saved as a draft. See Separating building from publishing.

Recommendations​

Start read-only. The List and Query tools cover most of what people actually want from this connector, such as asking which servers have no guards attached or which groups a user belongs to, and none of them can change anything.

Assign it to a narrow group. Connected groups control who can call it through the gateway, so scope it to admins rather than All Users. See Groups.

Attach a guard. Because the write tools mutate your governance configuration, this is a good candidate for a runtime guard that requires approval. See Guards.