AbuseZero MCP server and CLI: protect benefits from your agent
Set up free trial and credit abuse protection from your AI coding agent with the AbuseZero MCP server, or from your terminal and CI with the AbuseZero CLI.
You probably build with an AI coding agent already. The AbuseZero MCP server lets that agent set up and manage your abuse protection directly: create the benefit, pick a protection level, evaluate a claim without saving it, replay a policy change, investigate a denial. The AbuseZero CLI does the same from your terminal, your scripts and your CI.
Both give you the same actions as the AbuseZero REST API, with the same inputs, the same results and the same permission checks. This guide shows what you can do with each and how to connect them in a couple of minutes.
What your agent can do
| Task | AbuseZero MCP server tool | AbuseZero CLI command |
|---|---|---|
| Create a benefit to protect | abusezero_create_benefit | abusezero benefits create |
| List protection levels and policies | abusezero_list_policies | abusezero policies list |
| Evaluate a claim without saving it | abusezero_dry_run_evaluation | abusezero evaluations dry-run |
| Replay a policy change against past traffic | abusezero_replay_evaluations | abusezero simulations replay |
| Look up an evaluation and its reason | abusezero_get_evaluation | abusezero evaluations get |
| Add an allow or block list entry | abusezero_add_access_list_entry | abusezero access-lists add |
| Choose how a benefit treats automation and AI agents | abusezero_update_benefit | abusezero benefits update |
| Report an outcome | abusezero_report_outcome | abusezero outcomes report |
That's a sample. Every action that has an MCP tool also has a CLI command. The coverage table lists them all.
It's also the fastest way to respond when something goes wrong. When a wave of repeat claims hits at 2 a.m., one sentence to your agent can replay the last hour against a stricter protection level and switch the benefit over, without hunting through a dashboard half asleep.
A session with your agent
Prompts that work well:
Create a benefit called "Pro trial" with key pro_trial, Balanced protection,
once per user, in observe mode. Then add the evaluate call to our trial route.Replay the last week of pro_trial evaluations against the Strict protection level
and tell me which results would change.Why was evaluation 83647251920638517429 denied?The AbuseZero MCP server also offers ready-made prompts: review_policy, investigate_denial and simulate_policy_change. None of them changes anything on its own.
Connect the AbuseZero MCP server
Add a remote MCP server with this address. Any client that supports the standard Streamable HTTP transport can connect.
{
"mcpServers": {
"abusezero": { "url": "https://mcp.abusezero.com/mcp" }
}
}Your app opens a browser to sign in to AbuseZero, where you approve the scopes it asks for.
After you sign in, your agent lists your workspaces with abusezero_list_workspaces and picks one with abusezero_select_workspace. The full setup is in AbuseZero MCP server.
MCP access on every plan
The AbuseZero REST API and MCP server come with every plan, Builder included. Start with Builder and connect your agent today.
Built to be safe with an agent
Letting an agent change your abuse protection should feel safe. The AbuseZero MCP server is designed for it:
- Your scopes, never more. Sign-in uses OAuth. The agent gets only the scopes you approve, still limited by your role in the workspace.
- Every tool is labeled. Each tool is marked as read-only, changing data or permanent, so your app can ask before it acts.
- Retries never double up. Every change takes an idempotency key, so a retried call returns the original result instead of repeating the change.
- You hear about new connections. AbuseZero emails you when you approve a new app, and you can disconnect it at any time in Developers → OAuth apps.
- Keys, money and members need your say-so. They need the
operations:managementscope, which an API key alone never has. Your own sign-in and security, ownership transfers and deleting the workspace stay with you in the dashboard.
Every evaluation is real and uses credits, saved or not, whichever tool runs it.
Use the AbuseZero CLI in your terminal and CI
npx abusezero --help
abusezero login
abusezero workspaces list --output table
abusezero benefits listabusezero login opens your browser to sign in. On a machine without a browser, use abusezero login --device. For scripts and CI, create a machine client in Developers → OAuth apps and set its credentials as environment variables:
export ABUSEZERO_CLIENT_ID="…"
export ABUSEZERO_CLIENT_SECRET="…"
abusezero login --machine-clientMachine clients belong to the workspace, so your pipeline keeps working when people join or leave the team. See AbuseZero CLI.
From setup to production
The agent and CLI handle setup and operations. Your application still makes the evaluation call when a user claims a benefit, through the AbuseZero Node.js SDK or the REST API. The full flow is in add a trial and credit abuse check to your signup flow, and the playbooks for AI credits and API free tiers show the settings we'd start with.
Frequently asked questions
What is the AbuseZero MCP server?
It's a remote server that speaks the Model Context Protocol, so AI apps and coding agents can use AbuseZero's actions directly: creating benefits, managing policies and lists, evaluating claims without saving them, replaying changes and reading evaluations.
Which AI apps work with the AbuseZero MCP server?
Any client that supports the standard Streamable HTTP transport can connect to the remote server. Apps that only start local servers can use npx -y abusezero mcp.
Is it safe to let an agent change my abuse protection?
The agent works with the scopes you approve, limited by your role. Every tool is labeled read-only, changing or permanent, changes are idempotent, and keys, billing and members need the operations:management scope.
Can I use the AbuseZero CLI in CI?
Yes. Create a machine client in Developers → OAuth apps, set ABUSEZERO_CLIENT_ID and ABUSEZERO_CLIENT_SECRET, and every command uses it.
Does the MCP server cost extra?
No. REST API and MCP access come with every plan. Evaluations use credits wherever they run, saved or not.
Connect your agent
Start with Builder
Pay as you go. MCP access included.
AbuseZero MCP server docs
Connect, authenticate and choose a workspace.
AbuseZero CLI docs
Install, sign in and script every action.
Keep reading
- Add a trial and credit abuse check to your signup flow
- How to stop free tier abuse on an API product
- AI agents are claiming your free trials. Let the right ones in.
- Why we built AbuseZero: an abuse prevention API for SaaS
- AbuseZero webhooks: every evaluation in your own systems
- Vibe coded your SaaS? Don't wake up to a $6,000 bill