AI Agent for LinkedIn Ads That Can Make Changes

How PaidSync.ai keeps you in control

• Drafts first. A new LinkedIn campaign starts as a draft, which never serves and never spends, until you choose paused or active.
• Preview first. Creates run as a dry run by default and change nothing until a second, explicit call.
• Safe defaults, stated. LinkedIn Audience Network and audience expansion are off on new campaigns unless you ask for them.
• Approval before each change. In the PaidSync.ai workspace, every change waits as an approval card until a person approves it. From Claude, your assistant's own approval setting decides whether it asks first.
• A change log. Each live LinkedIn change is recorded with the user, the ad account, the tool and the outcome.
• Where changes run. LinkedIn changes run from Claude and other MCP clients. ChatGPT's app reports on LinkedIn today.

Is there an official LinkedIn Ads MCP server? (checked September 2026)

When PaidSync.ai checked LinkedIn's Marketing API documentation and API change log in September 2026, neither listed an MCP server published by LinkedIn. Every LinkedIn Ads MCP server in use is built by a third party on the Marketing API.

Why teams run LinkedIn Ads through PaidSync.ai

• It builds, not only reports. LinkedIn tools that make changes, including campaign groups, campaigns, Sponsored Content ads, conversion rules and lead gen forms.
• Checks before changes. New items start as drafts, most writes run as a dry_run preview, and destructive calls need confirm_destructive.
• B2B breakdowns. Performance by job title, company, seniority, industry, country and region.
• LinkedIn next to everything else. One connection reaches 14 platforms and 630+ tools, Google Ads and Meta included.
• What it does not do. It counts leads and cost per lead but does not read the leads themselves (names and emails) from lead gen forms.
• Price. PaidSync.ai is free for 15 tasks a month with no credit card and no time limit. Pro is $99 a month for 600 tasks, or $75 a month billed yearly, and Team is $249 a month for 5 seats and 6,000 shared tasks (pricing).

The same connection carries the Google Ads MCP server, the Meta Ads MCP server and the Microsoft Ads MCP server.

What an AI agent for LinkedIn Ads does

Three things in practice. It reads, so the assistant can pull impressions, clicks, leads, and cost per result without you exporting from Campaign Manager. It writes, so campaigns, campaign groups, ads, creatives, targeting, and conversions can be created or changed through conversation. And it gates, so most writes preview as a dry run first and destructive changes need an explicit confirm. The reading part is available on several LinkedIn MCP options. The depth of the write surface and the honesty of the gating are where they differ.

Every LinkedIn tool, by category

PaidSync.ai's live server exposes the LinkedIn tools below, by category (server total 630+ across 14 platforms). All 34 are named in the table below; the live count is also fully discoverable from any session via paidsync_context with include_tools:['linkedin_ads'].

Three of these carry the most operational weight. create_linkedin_campaign builds a full campaign against an existing campaign group, with objective, bidding type, daily or lifetime budget, targeting facets, and status set in one call. update_linkedin_campaign_targeting patches the targeting spec on a live campaign, job functions, seniorities, industries, company sizes, and member skills, without rebuilding the campaign. And get_linkedin_performance_by_dimension breaks performance down by campaign, creative, company, job title, industry, seniority, country or region, so cross-cuts that take clicks in Campaign Manager take one prompt here.

For marketers

Run LinkedIn Ads with an AI agent

A practical week looks like this. On Monday you ask what changed in the last 7 days across LinkedIn campaigns. The assistant pulls spend, leads, and cost per lead from get_linkedin_performance and gives you a plain-number answer. You spot an ad group targeting too broadly, ask for the targeting breakdown, review it, and ask the assistant to tighten seniority to Director and above. It proposes the update_linkedin_campaign_targeting call with the exact change shown as a dry-run preview, then applies it when you tell it to go ahead.

Because the same server carries Google Ads, Meta Ads, GA4, and GTM, the follow-up question "did the new LinkedIn creative produce GA4 conversions at a lower CPA" is answered in the same session, not in a separate tab. For teams running cross-channel demand generation, that session continuity is the practical difference between a LinkedIn connector and a paid-media server.

Write access brings an honest concern about safety on live ad accounts. The answer is procedural. Most LinkedIn write tools default to dry_run, which returns the planned payload without executing it, and image and video uploads apply on the first call. Live pauses and enables need an explicit confirmation flag. Over MCP, PaidSync writes no approval card, and your assistant's own per-tool approval setting is a second check you control. Sessions can be revoked at any time from the PaidSync connect panel. Read operations never require confirmation because they change nothing.

For agent builders

The protocol surface behind the tools

PaidSync is one MCP endpoint covering 14 platforms; architecture detail is on the MCP server page. To keep client context small, tools/list returns a lean starter set. The full catalog is discovered at runtime through paidsync_context, which returns the active account, connected platforms, and the grouped tool catalog. Pass include_tools:['linkedin_ads'] to get the exact LinkedIn tool names, signatures, and approval-gate defaults. Then execute any tool through paidsync_exec via await paidsync.call('tool_name', {...args}). That two-call pattern means an agent can enumerate and invoke every LinkedIn tool without a 630+ tool schema payload in every conversation turn.

The server-side checks run whatever the client. Most mutation tools default to dry_run for a no-op preview of the planned change, destructive operations need confirm_destructive, and sessions can be revoked without touching the OAuth client. For discovery and verification, the domain publishes OAuth authorization-server and protected-resource metadata (RFC 8414 and RFC 9728), an MCP server card, agent-auth.json, security.txt, and a Content-Signal directive, so an agent can establish what this server is and how to authenticate before its first task.

LinkedIn's Marketing API enforces rate limits. High-volume reporting loops will hit them, so design polling loops with a back-off and prefer dimensional breakdowns over per-campaign loops.

Connect LinkedIn Campaign Manager to your AI client

The endpoint is the same in every client, https://mcp.paidsync.ai/mcp. Sign-in is OAuth 2.0 with your PaidSync login. Once the client is connected, run connect_linkedin_ads to authorize your Campaign Manager account, then set_linkedin_ad_account to pick the account the session works on. The steps below are checked against each vendor's own docs.

Claude

In claude.ai or the Claude Desktop app, open Customize, then Connectors, click +, search PaidSync, click Connect, and sign in with your PaidSync login. Connectors work on every Claude plan, including Free, which allows one custom connector. Full walkthrough at connect LinkedIn Ads to Claude.

Claude Code

Run claude mcp add --transport http paidsync https://mcp.paidsync.ai/mcp. Start a session, type /mcp, pick paidsync, and sign in with your PaidSync login in the browser window that opens. For a team, commit a .mcp.json at the repo root with the same http server entry. Full guide: LinkedIn Ads in Claude Code.

ChatGPT

In ChatGPT, open Plugins, search PaidSync, click Install plugin, and sign in with your PaidSync login. If your workspace does not show the directory, open Settings, then Security and login, and turn on Developer mode, then in Plugins click the plus button, add https://mcp.paidsync.ai/mcp as a remote MCP server and choose OAuth. OpenAI says Developer mode is available to Pro, Plus, Business, Enterprise, and Education accounts on the web. Full walkthrough at connect LinkedIn Ads to ChatGPT.

Gemini CLI

Run gemini mcp add --transport http paidsync https://mcp.paidsync.ai/mcp, then /mcp auth paidsync and sign in with your PaidSync login. Gemini Code Assist and Antigravity read an httpUrl entry in ~/.gemini/settings.json. Full walkthrough at connect LinkedIn Ads to Gemini.

Copilot Studio

Open your agent, then Tools, Add a tool, New tool, Model Context Protocol. Paste the server URL, choose OAuth 2.0 with Dynamic discovery, transport Streamable HTTP. Click Create, then Add to agent, test, then publish to the Microsoft 365 Copilot channel. Full walkthrough at connect LinkedIn Ads to Microsoft Copilot Studio.

Perplexity

Open your profile menu, then All settings, then Connectors, then Custom connector. Name it PaidSync, paste the server URL, choose OAuth and Streamable HTTP, save, and sign in with your PaidSync login when prompted.

Cursor

Add {"mcpServers":{"paidsync":{"url":"https://mcp.paidsync.ai/mcp"}}} to .cursor/mcp.json in your project, or ~/.cursor/mcp.json for every project. Open Cursor Settings, then MCP, and sign in with your PaidSync login when Cursor prompts for OAuth.

A client that lets you set a header, such as Claude Code, can use an API key instead, sent as Authorization: Bearer ps_... or x-api-key. Never put the key in a URL; query strings end up in logs. Applying a change to Google Ads still requires the OAuth sign-in. On LinkedIn and the other platforms a key can apply changes too, so treat it like a password and prefer OAuth.

What this does not do

Stated plainly, because accurate scoping is more useful than overselling.

• Predictive Audiences beta is not exposed. LinkedIn's Predictive Audiences feature is not in the tool set. Targeting facets are supported through update_linkedin_campaign_targeting.
• Document Ads and certain vertical-specific formats are not documented. If Document Ads are central to your LinkedIn strategy, verify support with the PaidSync team before committing.
• No campaign deletion. PaidSync has no LinkedIn delete tool; pause a campaign to stop its spend.
• One object per task. Creating or updating 15 campaigns is 15 tasks or a structured loop in your agent. There is no single-call bulk mutator.
• No organic page posting or LinkedIn Pages management. PaidSync is paid media only.
• Rate limits apply. LinkedIn's Marketing API enforces rate limits, so high-volume automation loops should build in back-off logic.

Pricing

PaidSync is free for 15 tasks a month on all 14 platforms, with no credit card and no time limit. After that you pay for usage, the tasks your AI runs, not seats or a share of ad spend: Pro is $99 a month for 600 tasks, $149 for 1,200 and $199 for 4,000, or $999 billed yearly for 600, and you can add tasks if you run out. Team starts at $249 a month for 5 seats sharing 6,000 tasks. There is no per-account or per-platform charge. A task is one request from your AI assistant to PaidSync, such as reading a report or pausing an ad. Reads and writes count the same, and connecting an account or picking one uses no tasks. Full detail at paidsync.ai/pricing.

Connect LinkedIn Ads to your AI

One setup per client, same endpoint and sign-in everywhere.

• LinkedIn Ads to Claude
• LinkedIn Ads to Claude Code
• LinkedIn Ads to ChatGPT
• LinkedIn Ads to Gemini
• LinkedIn Ads to Microsoft Copilot Studio
• LinkedIn Ads to Perplexity
• LinkedIn Ads to Cursor
• LinkedIn Ads to Codex

Frequently asked questions