Agent keys — let any AI agent find, check and enrich records

Updated Markdown

An agent key lets an AI agent — Claude, ChatGPT, Clay, n8n or your own code — use ListPlus for your whole workspace, without first opening a list:

  • Find a person from a name and a company, a LinkedIn URL, a work e-mail or a question like "CFO of Tink in Berlin": name, job title, company, location, LinkedIn profile and how sure the match is. No e-mail or phone in this answer.
  • Check records the agent sends as JSON with the 100+ checks of the app (typos in e-mail domains, team mailboxes like office@, duplicates, phone codes that don't fit the country, LinkedIn status text in names …). Nothing is stored or changed; fixes come back as proposals.
  • Fill missing fields — work e-mail, mobile number, job title, LinkedIn profile, company data — with the same enrichment as in the app, charged per hit with the workspace's credits.
  • Verify e-mail addresses (deliverable or not).

Every answer lists what else ListPlus could do for those records, with the price in credits and in money, and a sentence the agent can pass on to you — so the agent asks before it spends.

For access to just one list (read, write, its own permissions), use that list's API connection instead — see "Give an AI agent access to one list".

Connect your AI in one click#

Open Agents in the left bar. The first time, the page shows one tile per AI: Claude Code, Claude (claude.ai), ChatGPT, Cursor, OpenClaw and n8n / Make / Code. Once something is connected, the same choice opens with Connect agent at the top right. ListPlus makes a key named after the tool and shows exactly the steps for it, with the key filled in — a command to paste, the server address and header to enter, or (Cursor) a button that installs it. Claude and ChatGPT carry their own logos. n8n / Make / Code gives the REST address, the HTTP header and the teaching URL (https://listplus.ai/a/lpk_…): paste that into any agent and it learns what ListPlus can do, with prices. Claude has a button, Add to Claude: it opens Claude's "Add custom connector" dialog with the name and address already filled in — click Continue, keep No sign in, click Add. The connector then works in Claude on the web, in the desktop app and on the phone. (The address holds the key; the safer way — the key as a request header — is folded below the button.) ChatGPT (tested with Pro): Customize → Plugins → Add → Add custom MCP server, paste the address shown, set Authentication to No authentication, tick the confirmation and click Create as plugin. In Business and Enterprise accounts an admin may first have to allow custom MCP servers.

The key is shown once. Treat it like a password: anyone who has it can spend up to its day budget. Only the workspace owner or an admin can connect an AI; members see the connections.

Your connections#

Under Connections each key has a row: what it spent today of its day budget and when it was last active ("never used" if no agent has called it yet). Manage (owner and admins) shows its totals, who made it and its agent list, and changes:

  • Name — e.g. "Clay" or "my Claude".
  • Budget per day — the most this key may spend in credits per day (default 100). With 0 the key spends nothing: person search and checks only.
  • Expires (optional) — after this day the key stops working.
  • Revoke stops the key at once. Its running jobs are cancelled — you pay only for what they found so far. Its agent list stays.

What it costs#

  • Person search and checks don't use credits; they have daily limits (with a key: 500 person searches and 1,000 records per check call).
  • Filling fields is charged per hit, at the prices of "Price list: what every action costs" — the same as in the app. Before an agent runs it, it can ask for the exact price of its records (estimate): "expected" and "at most". A record that already has a field costs nothing for it.
  • E-mail verification: 0.3 credits per address.
  • AI checks (check_records with ai: true, key only) find what rules cannot — a fake name, a person's name in the company field, a department instead of a job title. They are charged by the AI's real use, never above the "at most" the check's answer names.
  • Everything is paid from the workspace owner's credit balance, never more than the key's day budget, and never more than the agent's own max_credits for a call. Every answer tells the agent the balance and what is left of the day budget; when credits run short, it gets the address where you buy more, to pass on to you.

Agent activity#

Agent activity on the Agents page lists what your agents did in the last 7 days (at most 200 calls), by day: one row per task — "Records checked", "Person searched", "Cost estimated", "E-mail, Mobile number added" — with the time, the agent, how many records and what it cost. An enrichment that is still working says running and shows its real cost when it ends. A task the agent was refused says why (for example "day budget used up"). Above it, a heatmap shows the tasks per day of the last 91 days (darker = more; point at a day for its tasks and credits; a click on one of the last 7 days jumps to its tasks), beside it the credits of today and the tasks and credits of the 91 days. All agents narrows it to one connection. Status checks of a running job are not listed. The records themselves are not in this list — click "1 record" on an enrichment to open the key's agent list.

Where enriched records go#

Records an agent enriches go into the key's own list ("Agent · Claude Code") in your workspace. You can open it from the key's row on the Agents page and work on it like any list. Every record an agent gets back carries a link to it in ListPlus — the agent can hand it to you, and a click (signed in) opens the record. A record sent twice is the same row. The rows count toward your plan's record limit; when the list is full, the next call starts a new one.

Without a key#

Agents can try the person search and checks without any key at https://listplus.ai/a or the MCP server https://listplus.ai/mcp, with lower daily limits (30 person searches, 25 records per check). Hosted agents like ChatGPT or claude.ai share internet addresses with many users, so the limits without a key may already be used up — a key avoids that.

When a key stops working#

A key stops working when it is revoked or has expired, when the person who made it leaves the workspace, or when the owner switches off Agent API and MCP under Settings → Privacy & compliance → Workspace policy.