Resources/Comparisons

The best LinkedIn Ads MCP servers in 2026, compared by someone who built two of them

Daniel Popa·Founder, AdPlug·September 7, 2026·16 min read

There is no official LinkedIn Ads MCP server. LinkedIn has not shipped one, and as of 7 September 2026 every option on this page is third-party code sitting on top of the Marketing API. I know because I wrote two of them: the open-source server most guides link to, and the hosted one my clients' accounts run through. So this is not a neutral list. It is the list I would give a fellow media buyer who asked me which of these I would trust with a live budget, with every claim about a competitor linked to the page it came from.

Last verified 7 September 2026. Prices are what each vendor lists today. I re-check this page monthly and log what changed at the bottom.

The short version

Twelve servers in the table, one line each. The ones that did not make the table, and the sources for every row, are further down.

If you only want one line: hosted and you need writes, pick between AdPlug, PaidSync, Markifact, and Ryze AI. Free and technical, run my repo. Free and not technical, Kiin or AdPlug's free tier. LinkedIn numbers next to your other data, Two Minute Reports, Windsor.ai, Supermetrics, or Porter Metrics. I build AdPlug and I maintain the repo, so weigh that however you like.

Why there is no official server, and what that means for you

Google published its own Google Ads MCP server in October 2025. Meta ships one. LinkedIn has done neither, and I do not expect it to soon. The Marketing API is gated: you need a developer app, the app has to belong to a Company Page, you request the Advertising API product, and then you wait. LinkedIn does not publish a review timeline; the reports I have seen range from a few days to a few weeks. The Development tier you get first already allows unlimited reads and edits on up to five ad accounts you administer, which is plenty for one company or a small book of clients. Standard tier removes the account cap. AdPlug was granted Standard in June 2026.

That gate is why every server in this comparison falls into one of two families. Either you get the developer app yourself and run the server on your own machine, or a vendor holds the API access and you sign in through LinkedIn's OAuth screen and never see a developer portal. Hosted is also the only realistic way to reach ChatGPT, which wants a remote connector, and it is the only way to get a preview layer you did not build.

How I tested, and where I did not

Two of these ran against my own ad account: AdPlug, obviously, and the open-source repo, which is what I used before AdPlug existed. For everyone else I read the pricing page, the docs, the GitHub repo where there is one, and the tutorial videos, in the first week of September 2026, and I link the exact page under each name. I did not connect a client's account to a competitor to find out what it does, for the same reason I would not let a competitor connect to mine.

Two rules I kept to. Where a vendor's marketing page and its own documentation disagree, I say so; three of them have that problem. Where a number is not published, I write that it is not published rather than guessing.

AdPlug: the one I built because my own workflow was six Campaign Manager tabs

AdPlug exists because in early 2026 my LinkedIn work looked like this: pull a demographics report, export it, paste it into a chat, ask what to change, then go back and make the change by hand. The open-source server fixed the first half. AdPlug fixed the second half, and then kept going. On LinkedIn it exposes ten tools and 25 write operations: create, update, and delete campaign groups and campaigns; create ads and edit them in place across nine formats, single image, video, document, carousel, text, spotlight, message, event, and Thought Leader Ads; upload the media from the chat; build targeting and saved audiences and upload company or contact lists; create, update, and delete lead gen forms; and create conversion rules. A separate tool posts offline conversion events, up to 5,000 per call. Reads cover seven report types, targeting lookups with audience counts, and LinkedIn's public Ad Library, which needs no account at all.

Create a Thought Leader Ad in the Q4 Demand Gen campaign group from my latest post, daily budget $60, targeting marketing directors at SaaS companies in the UK. Preview it first; I will confirm before you execute.

All 25 of those write operations run as a preview first. You see the exact API call, you confirm, then a separate execute step runs it. The conversion-event upload is the one write that goes straight through, because it is a data push rather than an account change, and it draws on the same edit quota. Every tool call is logged with who, which account, which tool, the outcome, and when. Accounts start read-only. None of that is marketing-page decoration. In July I found that one of my own operations, the one that applies a targeting template to campaigns, was still calling an endpoint LinkedIn had retired, and the preview could not have caught it because it printed a label instead of the URL. Worse, the operation reported success with an empty results list. I reproduced both on my own account, fixed both, and previews now show the real endpoint for every operation. That is the argument for previews in one sentence: they only protect you if they show the actual call.

Pricing is deliberately generous on the axis agencies feel first: every plan, including Free, allows unlimited ad accounts. Free is read-only forever, with 500 requests in the first month and 50 a month after, which covers a proper audit. Pro is $29 a month for all platforms, 5,000 requests, and 200 previewed edits. Max is $89 with no meters. The same connector carries Google Ads and Microsoft Ads, so LinkedIn cost per lead against paid search is one question, not two exports. It works with Claude, including the connector directory, ChatGPT, Cursor, and Claude Code, and the setup guide is about two minutes long. What it is not: a toolkit for developers embedding ads data in their own agents, and there is no self-hosted version. For that, keep reading.

The open-source route: my repo, and the other one you will run into

I published danielpopamd/linkedin-ads-mcp in January 2026 because I wanted Claude Desktop to read my accounts and nothing else did. It has since grown to 25 tools: accounts, campaign and creative performance, demographics with real names instead of URNs, conversions, lead gen forms, period comparisons, and full create, update, and delete on campaign groups and campaigns. MIT licence, Node 18 or newer, tokens refresh themselves, rate limits back off themselves, and the README carries a prompt you paste into Claude Code so it does the install for you. The week I checked, Google's AI Overview cited it three times for this search, which is more than the hosted product gets.

What it costs you is the LinkedIn side. A Company Page to own the developer app, the Advertising API request, the callback URL registered in the app, and the wait. There is no preview layer, so if you switch the write tools on, the guardrails are whatever you build. I would not point it at a client's account without them.

The other repo you will run into is mharnett/mcp-linkedin-ads. Its README calls it production-grade, ships forty contract tests, and is careful about scope: seven tools, read-only today, with a write-mode gate already in place for the write tools it has not shipped yet. If you want a small, careful server, take it. If you want the full surface, take mine and add your own guardrails. Two warnings for the rest of the field. The npm package named linkedin-ads-mcpbelongs to AdKit, not to my repo; as of September it is a version 0.1.0 stub that describes itself as "initial release coming soon", and at least one popular tutorial tells you to install it. And one community server skips the developer app by scraping your logged-in browser session, which is the exception I flagged above.

PaidSync: dry run by default, and the widest net

PaidSync is the hosted server with the biggest footprint I found: 470-plus tools across fourteen platforms, with 25 LinkedIn Ads tools covering campaign groups, campaigns, creatives, conversions, targeting, and reporting. Its page says 25 in one place and 26 in another; either way it is the second-deepest LinkedIn write surface here. LinkedIn writes arrived in March 2026. Every write defaults to a dry run and waits for a human, deletes and pauses need an explicit confirmation, and there is an execution trail afterwards. It connects to Claude, ChatGPT, Gemini, Perplexity, Copilot, Cursor, and Windsurf, and its page carries a "last updated" stamp of 2 September 2026, which I wish more vendors did.

Pricing is metered by tasks: 15 a month free, Pro from $99 for 600 rising to $199 for 4,000, Team from $249 with five seats. If your team already lives on a dozen platforms and your monthly change volume is predictable, this is the most complete single connector on the page. The things I would model first: the task meter is the pricing axis, the free plan is small, and LinkedIn is one platform of fourteen rather than the focus. Its page also calls it the only MCP server with LinkedIn Ads write access. The table above disagrees.

Markifact: writes held for approval, inside an automation suite

Markifact comes at this from the automation side. Its LinkedIn page describes creating campaigns, ad sets, and ads in five formats, single image, video, carousel, text, and Spotlight, plus budgets, pause and activate, and renames. It attaches existing lead gen forms but does not create them, and it builds matched audiences from lists you already have. Every change is held until someone approves it, and execution logs keep for seven days on Pro and thirty on Team. The same server reaches fourteen ad platforms and a pile of analytics and workspace tools, runs in EU data centres, and connects to Claude, ChatGPT, Cursor, Codex, Gemini CLI, Windsurf, and Antigravity. It is the first citation in Google's AI Overview for this search, and its six-minute tutorial from August walks through connecting, reporting, and creating a campaign.

Pricing, from its pricing page rather than the LinkedIn page, which has none: 200 one-time credits free, Pro at $79 a month for one seat, Team at $249 for up to fifty. Pick it if you want scheduled, workflow-style automation and the MCP is one entry point among several. If you only want LinkedIn in a chat window, you will be paying for a suite you do not open.

Ryze AI: the flat-price newcomer

Ryze AI is a hosted remote server from Meow AI, over Streamable HTTP with OAuth 2.0, and its page was updated on 5 September, which makes it the freshest entry here. Reads include analytics with demographic pivots, campaign groups, campaigns, creatives, targeting facets, and lead gen forms with responses. Writes cover status, budget, bid, schedule, and targeting updates, creation of campaign groups, campaigns, and creatives, and delete or archive. The safety model is approval per tool, and deletes only run against an item-by-item approved list, with the page noting that deleted LinkedIn campaigns cannot be restored, which is true and which most vendors forget to mention. The same connector covers Google, Meta, TikTok, GA4, Search Console, and Shopify, 150-plus tools in total, and it lists Claude, ChatGPT, Claude Code, Cursor, Windsurf, VS Code, and Gemini CLI as clients.

Connecting the MCP is free. The $89 a month on the page is for a separate product, its Paid Ads Autopilot managed service, with a seven-day trial; do not read it as the price of MCP access. It is new enough that I would run it read-only for a month before turning the approvals on.

Adspirer: 54 LinkedIn tools, campaigns born paused

Adspirer has the largest LinkedIn tool count on the page: its open-source core lists 54 LinkedIn Ads tools next to Google and Meta, and new campaigns are created paused so nothing spends before you look. It is hosted, priced by monthly calls, 15 free, then Plus at $49, Pro at $99, Max at $199. Paused-on-create is a lighter guardrail than a preview: you review after the assistant builds, inside Campaign Manager. Fine for an operator who lives there anyway.

Kiin: free, LinkedIn-only, read-only, and clear about it

Kiin is the only other LinkedIn-first entry. It is a free MCP from a UK LinkedIn Ads agency founded in 2018. Its pricing page lists Kiin Intelligence at $99 per account per month and an Agency plan at $249 a month for three connected accounts plus $59 for each additional one, with a fourteen-day trial and no card. It says plainly that it "can't change campaigns, budgets, bids or targeting", which I respect more than the vendors who imply otherwise. What it does is analysis: named reports, rules, and playbooks benchmarked across the accounts it has connected, plus HubSpot attribution. It grew on founder-led LinkedIn posts rather than search, and if you want a free LinkedIn-only analyst inside Claude or ChatGPT, start there.

You outgrow it the day you also run Google, or the day you want the assistant to act on what it found. Per-account pricing on the paid tier is also the axis that hurts agencies fastest.

Quick takes: AdKit, Synter, Two Minute Reports, Zapier, CData

AdKit

AdKit calls itself "the safe MCP server for LinkedIn Ads", and the claim is concrete: every change lands as a draft you approve before LinkedIn sees it. Writes cover launching campaigns, building audiences, budgets, pause and relaunch, and targeting, across Meta, Google, TikTok, Reddit, X, and Microsoft too. No free tier: a single project is $29 a month billed yearly or $49 monthly, multiple projects $49 yearly or $97 monthly, seven-day trial, and no seat, per-agent, or percentage-of-spend fees. Its LinkedIn write depth is described in categories rather than counted, which is the one thing I would ask before paying.

Synter

Synter is a hosted server across 27 platforms, authenticated by API key. Its LinkedIn tools create campaigns, update budgets, pause, build matched audiences, and download leads. Two things before you sign: its blog and docs state no preview or approval step for writes, and pricing stacks a subscription, Solo at $20 a month or Scale at $500, with 3% of managed ad spend and metered actions such as $5 per campaign launch. On a real B2B budget that percentage is the number to look at.

Two Minute Reports

Two Minute Reports is a reporting product whose MCP listing sat at the top of Google's US results the week I checked. It is read-only and says so: Claude cannot create, edit, pause, or delete anything. Thirty-plus connectors, a spot in the Claude connector directory, no free plan, Lite from $81 a month billed annually with a fourteen-day trial. A good reporting bridge with an unusually good directory presence.

Zapier MCP

Zapier's MCP exposes seven LinkedIn Ads tools: create an audience, add companies or contacts to it, remove an email, create a report, and triggers for new lead gen form responses. No campaign, budget, or creative writes. The same connector reaches nine thousand apps, the free tier is 100 tasks a month, roughly fifty MCP calls, and Pro starts at $19.99. Use it to push lead forms into a CRM, not to run the account.

CData

CData is the enterprise data-connectivity company. Its hosted LinkedIn Ads MCP answers SQL-style questions the way its drivers query hundreds of other sources, and it publishes a read-only local server on GitHub. LinkedIn writes are not published anywhere I could find. Fourteen-day trial, then Standard at $79 a month billed annually. It is for data teams and IT, not for whoever is adjusting bids on a Tuesday.

The reporting bridges, and the ones whose pages and docs disagree

A whole tier of this market is marketing-data companies exposing the LinkedIn data they already pipe into dashboards. Supermetrics publishes a Claude directory connector from €39 a month billed annually. Its LinkedIn page says you can create and pause campaigns from Claude; its own campaign-management docs, dated July 2026, list Google, Meta, Microsoft, and TikTok and no LinkedIn. Treat it as read until the docs catch up. Windsor.ai has a free plan for one source and does allow pause, enable, and budget changes on LinkedIn, but with no server-side preview: its docs ask the chat client to confirm with you. Porter Metrics has a free plan, charges $12.50 per data-source account a month billed annually, never requests LinkedIn write scopes, and exposes six read-only methods, all of which it says outright. Its no-code walkthrough is the most-watched "connect LinkedIn Ads to Claude" video there is.

The strictly read-only group is refreshingly plain about it. Databloo, seven tools, a free tier for one account per connector, "read-only, always". Catchr, from $24 a month, no free tier. Adzviser, 80 metrics and 44 breakdowns, a 99-cent trial then from $48.99. BlueAlpha, ten LinkedIn analysis skills, no published price, writes on the roadmap. DemandSense, from the agency Impactable, joins ad exposure with website visitors and CRM records and is free during early access. GrowthSpree, free, Claude only. Two oddities: Insightful Pipe's hosted page lists ten write actions that create campaigns paused while its GitHub README still says read-only, and Improvado sells pause, budget, and targeting writes with contact-sales pricing and no stated safety step.

Reporting bridges are the right call when analysis is the whole job and LinkedIn is one source among many. They are the wrong call the moment the next step is a change to the account, and the ones with limited writes deserve the same questions as any write-enabled server.

The DIY route: Claude Code and the raw API

One more option is not a product at all. Anthony Blatner's tutorial at Speedwork connects Claude Code straight to the LinkedIn Ads API: create the developer app, request the read and write scopes, wire the OAuth redirect, then let Claude Code write its own API wrapper from LinkedIn's docs. I like it as a way to learn what the API can do, and a technical marketer ends up with a bespoke tool and no vendor. You also end up with no safety layer, no one to fix it when LinkedIn versions the API again, and the same developer-app wait as every self-hosted route.

What I would check before connecting any of them to a client account

If the accounts belong to clients, the comparison changes character. A misfired budget change on my own account is a bad afternoon. On a client account it is a lost client. LinkedIn adds two wrinkles of its own. The assistant inherits whatever Campaign Manager role the connecting profile has, so a profile with Account Manager rights on twelve client accounts hands all twelve to the server. And LinkedIn blurs its own hierarchy in everyday speech: what Campaign Manager calls a campaign, the API calls a campaign group, so "pause the campaign" can mean three different things. Before any of these touches client spend, get five answers in writing:

• OAuth, never shared logins or pasted access tokens. Access should be a scoped LinkedIn grant you can revoke from your own LinkedIn account.
• Read-only by default. Writes should be something you switch on deliberately, not the starting state.
• A preview or approval step on every account change that shows the real call and names the level, group, campaign, or creative, before it runs. No preview, no client accounts.
• A per-call audit trail. Who ran what, on which account, when, and whether it succeeded.
• Credential handling and paperwork. Encrypted token storage, a DPA you can countersign, and revocation from both sides.

That list is why AdPlug is built the way it is, and my own client accounts run through it, which is the strongest claim I can make. Credit where due: PaidSync's dry-run rule, Markifact's hold-until-approved gate, Ryze's per-tool approvals, and AdKit's drafts all clear the bar in their own way, and Adspirer's paused-on-create is a reasonable lighter version. The ones to keep away from client spend are the servers whose pages describe writes and no approval step, which today means Synter and Improvado, plus anything that leaves confirmation to the chat client, the raw open-source tools, and the DIY route.

How to choose

• Technical, willing to wait for LinkedIn API access, and you want no vendor in the loop: my repo, self-hosted, free. Or mharnett's if you want fewer tools and read-only today.
• You want a free LinkedIn-only analyst and will never need writes: Kiin, or AdPlug's free tier if you want unlimited accounts and the option to turn writes on later.
• You run accounts for a living and want safe writes with no setup: a hosted server with a preview or approval step. AdPlug if LinkedIn depth, Thought Leader Ads, and unlimited accounts matter and you also run Google or Microsoft. PaidSync if you live on a dozen platforms and task pricing fits. Markifact if you want automation workflows around the MCP. Ryze AI if you are on ChatGPT and want free MCP access with approvals. Adspirer if you are happy reviewing paused campaigns inside Campaign Manager.
• You mostly want LinkedIn data next to other business data: Two Minute Reports, Windsor.ai, Supermetrics, or Porter Metrics for analysis; Zapier if the point is moving lead forms into other apps.
• You want to learn the API itself: the Claude Code route, on an account you own.

Whichever you pick, the thing underneath is the same, and it took me the better part of a year of running accounts this way, Google first and LinkedIn since January, to say it plainly: the model matters less than the connection. An assistant with live Campaign Manager access and a preview gate beats a smarter assistant staring at last week's CSV, every week. If you want to feel that on your own spend, connect LinkedIn Ads and run the cost-per-lead-by-seniority prompt from the LinkedIn Ads and Claude deep dive. The read-only tier exists so you can judge this category on your own account rather than on anyone's table, including mine.

What changed since August

• Radiate B2B shut down; its MCP host stopped resolving. Removed from the table.
• The 19-tool open-source Campaign Manager server was deleted from GitHub.
• Ryze AI updated its LinkedIn page on 5 September and now lists creation writes.
• Markifact published its LinkedIn tutorial on 12 August, with campaign creation on camera.
• AdKit's npm package linkedin-ads-mcp is still a 0.1.0 stub; the tutorial pointing at it has not been corrected.
• Kiin's pricing page now lists an Agency plan at $249 a month for three accounts.
• For the record, not as evidence of quality: in the first week of September ChatGPT's web-search answer put mharnett's server first, and Perplexity named Ryze AI for ChatGPT users. Those answers move with the prompt, the day, and the model.
• Next check: first week of October 2026.