Background Agents
A background agent is an autonomous AI agent with its own identity, credentials, and a scoped set of capabilities. Unlike an interactive MCP client that a person drives, a background agent runs on a platform of your choice and calls tools through the Willow gateway using its own access key. Every call is subject to the same guards, policies, and logging as any other traffic.
Open it from Build > Background Agents.
Use background agents when you want an unattended workflow — a CI/CD bot, a support triage agent, a scheduled report writer — to have exactly the tools it needs and nothing more, with full auditability.
How a background agent is put together
Each agent has four parts:
| Part | What it is |
|---|---|
| Identity | A name, slug, description, and an access key / secret pair the agent uses to authenticate to the gateway. |
| Platform | The runtime the agent runs on (Custom, Claude Managed Agents, Claude Tag, AWS AgentCore, and more). |
| Capabilities | The tools, skills, and rules the agent is allowed to use. Capabilities are scoped per agent. |
| Triggers | Optional webhook events that start the agent automatically when something happens in a connected app. |
Because the agent authenticates as itself, its capabilities define its blast radius. Assign the minimum set needed for the job.
The list page
The Background Agents list is your organization's inventory of agents. The header has three actions:
| Action | What it does |
|---|---|
| Settings (gear icon) | Opens Background Agent Settings. |
| Refresh | Reloads the list and clears the search. |
| Create Agent | Opens the create dialog. |
Views
A toggle at the top right switches between two views:
- Table view: one row per agent, with search, advanced filters, and export. This is the default.
- Map view: a connector-centric grid. Each card is an MCP server, grouped by the agents that use it and the tools each agent has, color-coded by risk. Use it to answer "which agents can touch Salesforce?" at a glance.
Search and filters
The search box filters by agent name. Select Advanced Filters to narrow the list further:
| Filter | Filters by |
|---|---|
| Status | active or disabled. |
| MCP Server | Agents that have at least one tool from the selected server. |
| Risk Level | The highest risk among the agent's capabilities: Critical, High, Medium, Low, or Not Assessed. |
| Owner | Agents assigned to a specific owner. |
A badge shows how many filters are active, and Clear filters resets them all.
Creating an agent
Select Create Agent. The dialog offers three ways to start:
| Path | Use it when |
|---|---|
| Start from scratch | You want an empty agent and will add tools later. |
| Create with AI | You describe the agent in plain language and Willow selects the integrations, tools, and a system prompt for you. |
| From a template | You want a pre-built agent for a common workflow. Each template lists the connectors it uses and how many are already connected in your org. |
Create with AI
- Select Create with AI.
- In Agent Description, describe what the agent does. Be specific about the apps and actions it needs — the more concrete the description, the better the result.
- Leave Least Privilege on (default) to limit the agent to the minimum tools needed, or turn it off for broader access.
- Select Generate Agent. Willow analyzes your connected integrations and proposes a name, description, tool set, and system prompt.
- Review the preview. Regenerate to try again, or Continue to move to the details step.
AI generation requires connected integrations with tools. If Willow can't design an agent, refine the description and make sure the relevant MCP servers are connected.
Finishing the details
Whichever path you chose, the final step captures:
- Platform: the runtime for this agent (see Platforms). Managed platforms that are already connected are preferred and shown first; platforms that need a connection are disabled with a "Not connected" hint. Use Manage platforms to open settings.
- Name: the display name. The slug is auto-generated from it.
- Description: what the agent does.
Select Create. Willow creates the agent, adds any tools from the template or AI selection, and shows the agent's secret once in a toast — save it immediately, as it is not shown again. You land on the agent's detail page.
Platforms
The platform is the runtime the agent runs on. You connect platforms once in settings; individual agents then pick from what's available.
| Platform | Description |
|---|---|
| Custom | Bring your own runtime. Download the agent as a ZIP bundle or connect any host with the agent's MCP configuration. No org-level connection required. |
| Claude Managed Agents | Run the agent on Anthropic's managed infrastructure. Willow syncs the agent definition, tools, and gateway credentials to Claude. Requires an ANTHROPIC_API_KEY. |
| Claude Tag | Expose the agent's tools to @Claude as a connected app; Claude calls the tools over the REST API gateway. Skills and webhooks are not supported for this platform. No org-level connection required. |
| AWS AgentCore | Run the agent as an Amazon Bedrock AgentCore harness. Willow syncs the agent definition, gateway tools, and credentials to your AWS account. See Connect AWS AgentCore. |
| Codex, LangGraph | Marked "Soon" and not yet selectable. |
Managed platforms (Claude Managed Agents, AWS AgentCore) keep an external copy of the agent in sync. When an agent runs on one of these, its detail page shows a Sync action to push the latest definition, tools, and credentials to the platform.
The agent detail page
Selecting an agent opens its detail page. The header shows the agent's platform, status, a pending-request badge, and an Active / Disabled toggle. Managed-platform agents also show a Sync button. Work is organized into tabs:
Overview
Summary cards for tools, skills, triggers, and rules, plus recent activity (tool calls and webhook events) and the agent's connection details (MCP URL and API URL). Selecting a log entry opens the raw tool-call payload.
Tools
The tools this agent may call. Add tools from any connected MCP server, and remove them individually. Each tool carries a risk assessment and a policy check; Willow calculates missing assessments automatically when the page loads. This tab is also where you review and resolve capability requests from end users — a red badge shows how many are pending.
Skills
The skills assigned to the agent. Skills shape how the agent reasons; they don't call APIs. Add or remove skills, and re-run risk assessment per skill.
Triggers
Webhook-based triggers that start the agent automatically when an external event fires (for example, a new GitHub issue). Triggers use the agent's API URL.
Sessions
For Claude Managed Agents only: the conversation sessions the managed runtime has run, with their events.
Settings
Per-agent settings: owners, platform configuration (model, system prompt, credentials for managed platforms), connection details, and credential rotation.
Capability requests
When an end user opens an agent they own in the connect dashboard, they can request more tools, skills, or triggers. Each request becomes a notification tied to the agent. Pending requests appear as a red badge next to the agent name and on the Tools tab, where you can review the requested items (and any free-text explanation) and approve them by adding the capability, or dismiss them.
Background Agent Settings
Open Settings (gear icon) from the list header. Settings are split into cards.
Agent Platforms
Connect and configure the runtimes agents can use (API keys for Claude Managed Agents, AWS credentials for AgentCore, and so on). Only connected platforms are selectable when creating an agent. For AWS, follow Connect AWS AgentCore to set up the execution role and credentials.
Tool Assignment
| Setting | Effect |
|---|---|
| Allow common tools | Owners can assign tools from the organization's shared MCP servers. |
| Allow custom tools | Owners can generate and assign custom tools to their agents. |
Skill Assignment
| Setting | Effect |
|---|---|
| Allow common skills | Owners can assign skills from the organization's shared skill catalog. |
| Allow custom skills | Owners can create and assign custom skills to their agents. |
Credentials
| Setting | Effect |
|---|---|
| Show credentials to owners | Agent owners can view access keys and rotate secrets from the connect dashboard. Off by default. |
End User Permissions
| Setting | Effect |
|---|---|
| Allow end users to create agents | End users can create their own agents from the connect dashboard. Off by default. |
Select Save Changes to apply. These settings govern what agent owners and end users can do in the connect dashboard — the end-user view is documented in the user guide.
Enabling and disabling
Toggle an agent between active and disabled from the detail page header. A disabled agent keeps its configuration but cannot authenticate to the gateway or run, which is a fast way to pause an agent without deleting it.
Security notes
- The agent's secret is shown only at creation and when rotated. Store it in a secrets manager; it cannot be retrieved later.
- Capabilities are the agent's permission boundary. Prefer Least Privilege and remove tools the agent no longer needs.
- Every agent tool call flows through the gateway and is logged and subject to guards and policies, just like human traffic.
What to do next
- Assign tools from your MCP servers
- Review risk assessments
- Set up guards to constrain what agents can do
- End-user guide to Background Agents