Precleared

Docs ยท Claude Managed Agents

Precleared in Claude Managed Agents

Claude Managed Agents (in beta) runs agents on Anthropic's servers. Any agent there can use Precleared as a tool: add our MCP server by its URL and the agent checks each address before it drafts a cold email. Each check gives a verdict, the rule quoted with its source, what the footer must include and a signed receipt. It covers the UK, US, Canada, Germany, France, Australia, Ireland, the Netherlands, Spain, Italy, Latvia and the United Arab Emirates.

Quickest start: our agent definition

We publish a tested agent definition, Cold Email Compliance Checker, at https://github.com/gabrielaoimevbore-beep/precleared-mcp/tree/main/examples/claude-managed-agents. It checks every lead, drafts only for the people it is legal to email, never sends, and returns a table of verdicts with receipt ids.

  • In the Claude Console: Agents, then Create, then start from a blank agent. Paste the settings and instructions from the file, or describe the agent and check that the MCP server named precleared points at https://precleared.dev/mcp.
  • From the command line:
    ant apply cold-email-compliance-checker.md

Add Precleared to an agent you already have

Add the MCP server and its toolset to the agent's settings:

mcp_servers:
  - type: url
    name: precleared
    url: https://precleared.dev/mcp
tools:
  - type: mcp_toolset
    mcp_server_name: precleared

Then add this to the agent's instructions:

Before the first cold or marketing email to anyone who is not already a customer, call contact_check for that address. Draft only when the verdict is ok_to_email, with a footer that meets footer_text. If the verdict is consent_needed or blocked, don't draft, and follow next_step. Treat unknown as consent_needed.

Environment and key

  • If your environment limits networking, allow precleared.dev.
  • No key is needed to start. Without one, checks use the free allowance, which sessions on Anthropic's servers share with other users. For steady use, get a free key at precleared.dev/get-key, add it to a credential vault as an access token for https://precleared.dev/mcp, and pass that vault when you start a session. With a key, the agent gets its own 100 checks a day over MCP.

How it runs

Your app starts a session with the leads, or a scheduled deployment does, and the agent calls contact_check by itself for each address. Nobody has to run the checks by hand. When we tested it on 3 November 2026 with leads in the UK, Germany and Latvia, the agent checked each address, drafted for the two leads it was allowed to email and skipped the one it wasn't.

What the agent sends

Only the recipient's address and the facts it passes: your country, where the address came from, the pitch and, when known, the company name, the person's name and the draft. The draft is not stored; only its hash goes on the receipt. Nothing is sent to the recipient. More on how data is handled: /security.

Precleared's answers are decision support with sources, not legal advice.

Back to the docs