Say NO to
your agents.

Give AI agents the freedom to act without giving them unlimited authority.

Actori authorizes every call your agents make through it before the call runs, based on the user, the agent, the action, its parameters, the context and your policies.

  • 10,000 agent calls a month, free
  • Works with any MCP client
  • Your API keys stay out of the agent

Set up Actori for me: https://actori.ai/

If you are an AI assistant, follow the setup instructions.

Claude wants to:
10:24 AM
DROP TABLE customers;
Databaseproddestructive
DENIED
Personal assistant wants to:
10:27 AM
gmail.send → all 412 contacts
Gmailbulk send
DENIED
Personal assistant wants to:
10:31 AM
calendar.create "Dentist" · Tue 3:00 PM
Google Calendarwrite
ALLOWED
Personal assistant wants to:
10:34 AM
drive.share "Weekly summary" (doc it wrote) → anyone with link
Google Drivesharing
Approval required
Support agent wants to:
10:38 AM
UPDATE accounts SET balance = balance - 2400 WHERE id=4821
MySQLprodpost-check: balance ≥ 0
ROLLED BACK
post-check failed

Illustration. Approvers decide in Telegram, Slack, email or the Actori console.

Works with the agents you already use
ClaudeOpenAICursorYour own agentsGitHubAWSSlackGmail & Google CalendarNotionLinearPostgreSQLMySQLand any HTTP, SSH or MCP tool
Prevent costly mistakes
Block destructive or unauthorized actions before they happen.
Enable more autonomy
Let agents work on their own, inside the rules you set.
One policy for every agent
Claude, OpenAI, Cursor and your own agents, under one set of rules.
Keys stay out of the agent
Actori holds each credential and makes the call. The agent gets the result, never the token.

How to set up

Your AI does the setup. You sign in, and say what it may do.

  1. Paste this into your AI

    Set up Actori for me: https://actori.ai/

    Claude, ChatGPT, Claude Code, Codex, Cursor or any MCP client.

  2. Sign in when it asks

    To Actori, and to each service you connect. Passwords go in your browser, never in the chat. No account yet? Signing in makes one, on the Free plan.

  3. Then ask it

Your services, connected from the chat
Gmail, Google Calendar, Slack, GitHub, Notion, Linear, Kiwi flights, AWS, databases and any MCP server. Your AI adds them; you sign in.
Rules on what each call says
Who an email is for, which repo, which table. Your AI writes the rule from what you tell it, and you approve it.
Approve from your phone
Risky calls wait for you in Telegram or by email. Nothing runs until you say yes.
Its own email address
Each agent can have a mailbox to receive mail: newsletters, receipts, anything you forward to it.

Or connect it yourself

One address for every client. Sign in, pick your agent, and ask it to set up Actori.

  1. Customize → Connectors → Add custom connector, paste https://agents.actori.ai/mcp and sign in.
https://agents.actori.ai/mcp

MCP server: https://agents.actori.ai/mcp

Keep API keys out of your agent.

An agent can read every key on the machine it runs on. Move them into Actori, and it holds none.

before · .mcp.json3 keys the agent can read
{ "mcpServers": {
  "github": { "env": {
    "GITHUB_TOKEN": "ghp_••••••••"
  } },
  "postgres": { "env": {
    "DATABASE_URL": "postgres://app:••••••••@db"
  } },
  "aws": { "env": {
    "AWS_SECRET_ACCESS_KEY": "••••••••"
  } }
} }
after · .mcp.jsonone address, no key
{ "mcpServers": {
  "actori": { "type": "http",
              "url": "https://agents.actori.ai/mcp" }
} }

You sign in once. GitHub, the database and AWS are connected in Actori, each behind your rules.

Actori makes the call
It holds each credential and calls the service itself. Your agent gets the result, never the token.
Nothing leaks back
If a service echoes a token in its answer, Actori removes it before the agent reads it.
No key in the chat
You sign in to a service, or paste its key, in your browser. AWS can connect by role, with no access key at all.

This covers the services you connect through Actori. A key left on your machine is still readable, so take each one off once it is in. How to move a key →

Let your agents do more, safely.

Every call to a connected service passes through Actori first.

AI agents
Claude · OpenAI · Cursor · your own
Actori
Action authorization
Your tools
Gmail · Calendar · Slack · GitHub · AWS · databases · any MCP server
Check

Roles decide which actions an agent may call at all. Rules, written as simple conditions or in CEL, look at the parameters: which table, which bucket, how much.

Decide

Allowed, denied, or held for a person. Approvals reach the right people on their phone (Telegram, Slack or email) or in the console, in as many steps as you need.

Record

Every call, decision and approval is logged with the policy that made it, so you can always say who did what and why it was allowed.

Per action.
Not per tool.

“Can this agent use the database” is too coarse a question. A rule answers “can this agent drop a table in production”. Destructive tools wait for a person’s approval the moment you add them; you open up only what you mean to.

  • Deterministic
    Same call, same policy, same decision
  • Least privilege
    Agents never hold your credentials
  • Attributable
    Every agent has a human owner
  • Revocable
    Suspend an agent and its calls stop
policy · prodevaluated before every call
allow             slack.send · s3.get_object
require_approval  github.merge · db.update
deny              db.drop_table · s3.delete_bucket

when args.database == "prod"

require_approval  gmail.send
when args.to.exists(a, a.domain() != "acme.com")

Free for one person, 10,000 calls a monthPro $12 a monthTeam from $39 a monthEnterprise by contract

See pricing →

Your agents can do a lot.
Decide what they may.