Skip to content
Bring your own agent

LinkedIn for AI agents: drive a full outbound department from your own agent

Your agent can run LinkedIn and email outbound through Zevari. Connect Claude, Codex or ChatGPT over MCP, or OpenClaw, Hermes Agent or your own code over REST, and it drives the same six specialists the Zevari app uses: prospecting, signals, ICP scoring, voice writing, follow-up and reply qualification. A human approves every send. No browser extension, no cookies.

Connect your agentOr let the team run it
terminal

Over MCP

$ claude mcp add --transport http zevari https://mcp.zevari.ai/mcp
# then run /mcp in Claude Code and sign in

Over REST (bearer token)

$ curl https://api.zevari.ai/v1/search/posts \
  -H "Authorization: Bearer $ZEVARI_API_KEY" \
  -d '{"query":"hiring an SDR","posted_within_days":30}'

One department, two ways in

The specialists do the work. Your agent can direct them.

Zevari is an AI SDR team for LinkedIn and email. The same six specialists sit behind the app, the MCP endpoint and the REST API, so nothing is lost when you switch between them.

Prospecting

Runs saved searches and lands net-new people on your lists.

Signals

Finds LinkedIn posts by keyword and people engaging with yours.

ICP Scoring

Scores every lead against your ideal customer, with reasons.

Voice Writer

Drafts outreach in your voice, trained on what you have written.

Follow-up

Runs the sequence across LinkedIn and email, inside the limits.

Reply Qualifier

Sorts replies and flags the conversations worth answering.

Over MCP

For Claude, Claude Code, Codex, ChatGPT and other MCP clients. Add one endpoint, then sign in with OAuth or a bearer token. Commit the config to your project and the whole team gets the same connection.

{ "mcpServers": { "zevari": {
  "type": "http",
  "url": "https://mcp.zevari.ai/mcp"
} } }
Over the REST API

For OpenClaw, Hermes Agent, a cron job or anything you wrote yourself. One bearer token over plain HTTP, no MCP client required. Zevari holds campaign state, so a sequence keeps moving between your agent's runs. The developer docs have the full reference, and the developer page covers wiring it into an existing stack.

Works with the agent you already run

Pick your agent

Your agentConnect viaGuide
Claude / Claude CodeMCP (OAuth)Claude guide
CodexMCP (bearer token)Codex guide
ChatGPTMCPChatGPT guide
Other MCP clients (Grok Bot in Cursor and more)MCPCompatibility table
OpenClawREST (bearer) or MCPOpenClaw guide
Hermes AgentMCP server entry or RESTHermes guide
Your own agent, cron job or scriptREST (bearer)API reference

If it can call an HTTP endpoint, it can work through Zevari. Teams that pair contact data with LinkedIn work can start from MoltSets for LinkedIn.

Your agent prepares. A human approves the send.

An agent that can send on LinkedIn unattended puts the account and your name at risk. With Zevari, your agent can research, score, draft and sequence as much as it likes, and every write stops for your approval, in the app, in chat or in a digest, until you switch a trained specialist to Autopilot. Will it get your account banned? Not if Zevari is the only tool running outbound on the account and you use it lightly yourself while it works: on those terms we guarantee it. Here is what is enforced:

  • Every write, whether a message, connection request, comment or post, is staged for a human to approve. An agent can prepare it; it cannot send it unattended.
  • Hosted connector with a session-based connection. No browser extension, no cookie injection, no password handoff.
  • Weekly connection-request ceilings enforced in code: 150 on Free*, 150 on Premium, 200 on Sales Navigator.
  • * Free accounts: LinkedIn caps connection requests that carry a note at 3 a month. The weekly ceiling covers requests sent without a note, and Premium and Sales Navigator accounts get a far larger allowance for requests with a note. Most outreach uses a note, so Premium or Sales Navigator is the practical choice for real outbound - the full explanation is on /safety.
  • A daily cap on all account actions (400 Free, 1,000 Premium, 2,000 Sales Navigator), plus burst caps per minute and per hour.
  • Actions run inside your working hours and are spaced the way a person would space them.

What comes out the other end is measured as qualified conversations, not replies or a meetings number.

Read the safety model

FAQ

LinkedIn for AI agents, answered

Can I give my AI agent access to LinkedIn?

Yes. Connect your agent to Zevari over MCP (Claude, Claude Code, Codex, ChatGPT or any MCP client) or over the REST API with a bearer token (OpenClaw, Hermes Agent, or your own code). It can then drive the same six specialists the Zevari app uses: find prospects, score them, draft in your voice and run sequences. Every send still waits for a human to approve it.

Should I connect over MCP or the REST API?

Use whichever your agent speaks. MCP is the shortest path for Claude, Codex, ChatGPT and other MCP clients: one endpoint plus OAuth or a bearer token. REST suits self-hosted runtimes like OpenClaw and Hermes Agent, cron jobs and your own code: one bearer token over plain HTTP. The approvals and limits are the same either way.

Will an AI agent get my LinkedIn account banned?

No. If Zevari is the only tool running outbound on your LinkedIn account, and you use the account lightly yourself while Zevari is working, we guarantee it. Zevari's Safety Center paces every action like a person and counts every action the account takes against its limits: 100 connection requests a week on Free* accounts, 150 on Premium and 200 on Sales Navigator, daily caps, working hours and burst limits. Anything that would cross a limit waits for the window to reset. If another automation tool runs on the same account, or you use it heavily by hand at the same time, those actions fall outside Zevari's count, and we cannot guarantee it. The weekly figure covers connection requests sent without a note: LinkedIn caps requests that carry a note at 3 a month on a free account, and gives Premium and Sales Navigator a far larger allowance, so Premium or Sales Navigator is the practical choice for real outbound (see /safety).

I run a self-hosted agent on a VPS. Does this work?

Yes. Your agent keeps running where it is and calls Zevari with a bearer token. Zevari holds the campaign schedule and the approval queue, so a sequence keeps advancing between your agent's runs. See the OpenClaw and Hermes guides or the API reference.

Do I need my own agent to use Zevari?

No. Most customers let the six specialists run in the Zevari app and review what they stage. Connecting an agent is a second way in to the same team, for people who already work in Claude, Codex or their own code.

What does it cost?

One plan: $87 a month, or $497 a year (save 52%). It includes all six specialists and one LinkedIn account; each extra LinkedIn account is $37 a month, and extra users are free. First-time accounts get a 7-day free trial, and there is a 30-day money-back guarantee after the first charge.

Same team, whichever way you come in

$87 a month, or $497 a year (save 52%). All six specialists, one LinkedIn account, extra LinkedIn accounts $37 a month. 7-day free trial, 30-day money-back guarantee. See pricing.

Connect your agent

Add the MCP endpoint or create an API key, and drive the six specialists from Claude, Codex, ChatGPT, OpenClaw, Hermes Agent or your own code.

Open the quickstart

Let the team run it

Tell Zevari who you sell to. The specialists find the people, write in your voice, follow up and sort the replies in the app. Train them first, then switch them to Autopilot.

Build my AI team

Zevari: a full outbound department for LinkedIn and email, running today.