Linked API Overview

Linked API is a LinkedIn automation API that runs every action inside a dedicated cloud browser on an account you connect. One REST surface, a Node.js SDK, a CLI, an MCP server and n8n and Make nodes all drive the same 58 documented actions. Pricing is published and per seat, at $49 or $74 per connected account per month.

What Linked API is

Linked API sells one thing: programmatic control of a LinkedIn account its customer owns. The vendor’s own line for it is “Secure automation API for LinkedIn”. It writes for a developer or an AI agent, and the unit of everything is a seat, which holds exactly one connected LinkedIn account. It runs on LinkedIn and on no other channel.

The design decision underneath all of it is that a call is a workflow. A caller posts a workflow made of actions to one endpoint, each action can carry child actions through a then parameter up to five levels deep, and the result comes back as one JSON tree. Sending a connection request, commenting on a post and pulling a company’s employee list are the same request shape with a different actionType.

One disambiguation, because two vendors share a name. This page is about Linked API at linkedapi.io. LinkdAPI at linkdapi.com is a different company with a different product, a credit-priced data API that connects no LinkedIn account at all. Listicles and AI answers merge the two routinely, which gets the account model backwards in both directions.

The vendor states its position toward the platform in its own footer: the product is “not affiliated with, authorized by, endorsed by, or sponsored by LinkedIn Corporation”, and the trademark is used “only to describe compatibility”. That is the footing every account-based API in this category stands on.

How an account connects and a call runs

An account is connected through a hosted session. The customer’s server asks for one, gets back a short-lived link, and the person who owns the account opens it and logs into LinkedIn there. The account then appears in the Admin API with an id and its own identification token, and every call about it carries two headers: the workspace token and that account’s token.

Seats gate the connection itself. Asking for a session when every seat is held by an active account or a pending one returns noAvailableSeats, so adding an account is a billing action before it is a technical one. Each account then reports as active, frozen or reconnection_required, and the last of those arrives with a link to hand back to its owner, which is the day LinkedIn ended the session and a human has to sign in again.

Execution is deliberately unhurried, and the docs say so first. Because the platform drives a real browser, “requests are not executed instantly”: visiting a person’s page and liking their latest post “might take around 20 seconds”, and a company retrieval with an employee list “can take several minutes”. Workflows are accepted at once and queued, and they “will not be executed in parallel” on one account. The vendor states the ceiling that follows in its own words: “an account is only able to run one workflow at a time, so a window of a couple of hours per week puts a hard ceiling on how much work can drain through it.”

Working hours are a per-account schedule, and they change what a caller can assume about timing. Each connected account can carry a weekly activity window with an explicit timezone, and a request arriving outside it meets one of three policies.

• Queue until the window opens. The workflow is parked and starts by itself when the window reopens. This is the default for a newly connected account, and a parked workflow that has waited seven days fails with workingHoursWaitExpired.
• Reject with an error. The request fails immediately with outsideWorkingHours and nothing is queued.
• Run anyway. The caller’s own requests run at any hour. The vendor’s scheduled work does not: inbox and network synchronisation stay inside the window under every policy, and “the cloud browser behind the account is released outside the window under every policy too”.

The reasoning is published alongside it. An account that acts around the clock “looks like a user who never sleeps, which is one of the strongest automation signals there is”. The window is set in the platform by the account owner, and there is no API for changing it.

What it costs

Pricing is public, flat and per seat. Core is $49 per seat per month and Plus is $74, both billed against a seat that holds one connected LinkedIn account. Annual billing is marked “Save 29%”. A 7-day free trial needs no credit card. The two plans differ on one line: Plus adds Sales Navigator actions.

What a seat means is stated without hedging: “One seat equals one connected LinkedIn account. If you automate 3 LinkedIn accounts, you need 3 seats.” There is no published volume bracket. The pricing page ends the tier list with “Need volume discounts or a custom plan?” and points at the platform, and invoicing is offered on annual plans with five or more seats.

The consequence is arithmetic a buyer can do in their head, and it is worth modelling before anything is built. Ten connected accounts on Core is $490 a month. Fifty is $2,450. A hundred on Plus is $7,400. The published price does not bend at any of those points, so whatever a volume conversation yields sits outside what the site says.

Unlimited workflow execution is a billing statement, worth reading as one: no per-action fee and no credit counter on the vendor’s side. LinkedIn’s own weekly caps still apply to the account whatever the API charges for, which is what the next section is about.

The limits it enforces, and the limits it publishes

Two different things carry that word here. Linked API enforces caps of its own, server-side, in eleven action categories over daily, weekly and monthly periods, and a call that would break one returns limitExceeded without reaching LinkedIn. Separately it publishes a long guide to LinkedIn’s own limits, presented as guidance and not as platform fact.

The enforced half is an API of its own. Six documented calls read the system defaults, read an account’s limits, read usage against them, set them, delete specific ones and reset to defaults. The categories cover profile and company page views, connection requests, messages, search queries, reactions, comments and posts on the standard interface, plus Sales Navigator profile views, company page views and messages.

Defaults ship with a new account. The sample response in the reference returns stMessages at 50 a day and stConnectionRequests at 10, both enabled, and that is the only place a specific default number is published. Every value is the caller’s to raise, lower or switch off, and the vendor puts the responsibility in one line: “It is your responsibility to understand and follow the limits appropriate for your account.”

The published half is a reputation model. The limits guide sorts accounts into three levels, New, Established and Trusted, and gives a pace for each. It opens by saying where the numbers come from: “While LinkedIn doesn’t publish official limits, this guide combines community knowledge and our hands-on expertise.”

It carries more besides: page views by subscription tier, likes and comments by reputation level, the Commercial Use Limit on search at 300 queries a month for Free and Premium Career, per-query result ceilings of 1,000 profiles on standard search and 2,500 on Sales Navigator, and a 30,000 cap on total connections. It then grades itself, saying the guide “should be used as a reference point rather than official guidance, as actual limits vary by account and may change over time”.

That caveat is the honest one, because LinkedIn publishes no numeric invitation cap of its own. These figures are one vendor’s reading of what accounts survive, which is what every limits table in this category is. What LinkedIn does document is on our own LinkedIn connection limits page.

The API, the MCP server and the skills

One workflow endpoint, 58 documented actions behind it, and five ways to reach them: REST, a Node.js SDK, a CLI, an MCP server and nodes for n8n and Make. The actions split into a standard namespace of 46 prefixed st. and a Sales Navigator namespace of 12 prefixed nv., which is where the Plus plan earns its extra $25. Webhooks push workflow, account, inbox and network events, so a caller does not have to poll.

• Outreach and network. Send and withdraw connection requests, check status, retrieve pending requests, accept or ignore incoming invitations, list and remove connections, and monitor the network so acceptances arrive as events.
• Messaging and inbox. Send a message or reply into a thread by threadId, sync one conversation or the whole inbox and poll a single endpoint for everything arriving across it, and archive, star or mute a thread. Sales Navigator carries its own inbox and message action.
• Search and retrieval. People, companies and jobs with filters, a customSearchUrl escape hatch taking a LinkedIn search URL as written, and page fetches whose sub-actions reach experience, education, skills, employees, decision-makers and comments.
• Engagement and signals. React and comment on posts, react and reply to comments, open a person’s page so the visit registers, and read the account’s own SSI, analytics, feed and profile viewers.

The composition model does the work a batch endpoint does elsewhere: one workflow can search companies, retrieve employees for each result, then check connection status and send a request to each of those, as one nested call with one result tree. There is no campaign dashboard and the vendor does not sell one, so where a team wants a canvas the n8n and Make nodes are what the product offers, and how those platforms handle LinkedIn is on our LinkedIn integrations page.

The MCP server exposes those actions themselves, so an agent connected to it sends the connection request. Fifty-nine tools are documented across the standard interface, Sales Navigator, workflow escape hatches and account administration, and the source is published: “Source code available on GitHub.”

The administration block is the part to read twice before wiring it up. An agent holding the full toolset can set the number of subscription seats, connect a new account, disconnect an existing one, and rewrite or reset the rate limits on any account in the workspace. Those are billing and safety controls sitting in the same list as fetch_person, and an MCP client hands the model whatever the server advertises.

Two documented setups, carrying the credential differently. The hosted server at mcp.linkedapi.io is added as a custom connector, with the workspace token and the account’s identification token written into the connector URL as query parameters. The local option is an npx process, @linkedapi/mcp, taking the same two tokens as environment variables. Documented clients are Claude, Claude Desktop, Claude Code, ChatGPT, Codex, Cursor, VS Code and Windsurf, and one instance addresses one account.

Agent Skills are the newer surface and they are honestly labelled. One command, npx @linkedapi/skills, installs prewritten LinkedIn workflows into Claude Code, Codex, Cursor or Windsurf. Two are live, LinkedIn Growth and LinkedIn General. Three more carry a “Coming soon” badge and are not shipped: auto-outreach, job search and social selling. What changes when an agent drives a LinkedIn account directly is the subject of our LinkedIn MCP server page.

What the vendor says about account safety

The safety case here is architectural and documented in detail: one real browser per account, a matched device profile, a stable residential IP, actions paced like a person, and caps that stop a call before it runs. What the vendor does not publish is an outcome. No ban rate, no percentage, no sample and no measurement method appears on the safety page, on the site, or anywhere in its documentation.

• A dedicated browser, stated as such. “Each LinkedIn account gets its own full browser in the cloud”, described on the same page as “not headless emulation, not an HTTP scraper”, one that “renders pages, executes JavaScript, and behaves like the real browser you’d open on your own computer”.
• A matched device profile on a fixed address. “We maintain a stable residential IP from your region and match your complete device fingerprint”, naming operating system, browser version, screen resolution, language and timezone, with “+50 more parameters matched” claimed and unenumerated.
• Human pacing, with a worked timeline. “Every action executes at realistic speed with natural pauses”, and the page prints a nine-second sequence for one connection request, from opening the profile to clicking send.
• Caps that hold, once set. “When a limit is hit, the system stops and returns an error. It never exceeds your configured caps.” The caps in that sentence are the caller’s, and the dashboard values printed beside it are illustrative.
• Credentials the vendor says it never holds. “We never see, intercept, or store your password”, because sign-in happens inside the isolated browser, after which “no one, including our team, can reach that session”.

The one question a buyer actually asks is answered on the same page, and the full wording is the version to read. Asked “Can my account get blocked?”, the vendor answers: “Most account issues come from spam-like behavior or exceeding LinkedIn’s limits. Linked API is built to prevent exactly that: human-like pacing and rate limits you set for every action type keep your activity in a safe range.”

Read it for what it claims and what it leaves open. It names the two causes it addresses, attaches a mechanism to each, and stops there. There is no absolute in it, and no figure either. The absence is checkable in one pass, because the vendor publishes its whole documentation as a single 1.19 MB file and that file contains no ban rate, no safety percentage and no account count. So this is a case a buyer audits by reading the architecture and running it, resting on caps that are configuration: the defaults ship at 10 connection requests and 50 messages a day, and nothing argues with a caller who raises them.

The shape of job it fits

Linked API fits a developer who wants a LinkedIn account under program control and is happy to own the pacing decisions that come with it: a few connected accounts, a workflow composed in code, caps set per category, and a terminal or an agent as the interface. Where those four line up, this is a clean fit and the price is easy to model.

• A small account count. The seat price is flat, so one to five accounts runs $49 to $370 a month and the bill stays legible. The same flatness is what makes a hundred accounts a different conversation.
• Source you can read. The MCP server is published on GitHub under an MIT licence, which matters to anyone whose security review has to see what a tool sends before it is approved.
• Sales Navigator as a first-class namespace. The nv. actions are separate calls with their own inbox, search and fetch, so a team living in Sales Navigator gets it as a surface for $25 a seat.
• Pacing as your decision. Eleven categories, three periods and an API to change any of them, so a team with its own view of what an account can take gets to encode it.

The edges come out of the same design. One workflow at a time per account bounds throughput by wall-clock time, and a narrow working-hours window tightens that bound. The per-seat price does not bend at volume on any published page. And LinkedIn is the whole product, which is the right answer for some teams and a hard stop for anything that needs a mailbox too.

The other shape in this category is the same channel under a different contract. That is what GTM API is, and it is our own product: LinkedIn only, 160+ typed tools across 11 LinkedIn toolsets out of the 18 on the whole server, a preview-then-confirm step in front of an action, limits applied server-side before each call, and a free plan ahead of per-account pricing that falls as accounts are added. On that stack, gtm-api.com reports 20,000+ accounts at under a 1% ban rate. Email, messengers and calendars are on the roadmap there and are not shipped. Unipile takes the multichannel route across mail, messaging and calendars, and ConnectSafely.ai sits on the LinkedIn-only side of the line as well.

Two questions separate them, and neither is a feature list. Do you want a workflow you compose and cap yourself, or typed tools with the pacing decided for you? And how many accounts does the bill have to carry?

Frequently Asked Questions

Sources & Further Reading

• Linked API, product (“Secure automation API for LinkedIn”, the cloud-browser flow, the client list) · pricing ($49 and $74 per seat, “One seat equals one connected LinkedIn account”, the trial, the volume-discount line, the affiliation disclaimer) · summary for answer engines (the documentation map)
• Linked API, safety (every quotation in the safety section, the device-profile parameters, the action timeline and the credential flow) · understanding LinkedIn limits (the reputation tiers, the pace tables, the Commercial Use Limit and the closing caveat)
• Linked API developer docs, core concepts (workflows, request timing, sequential execution, the responsibility line) · actions overview (the 58 actions, the limitExceeded error) · working hours (the weekly window, the three policies, the one-workflow-at-a-time ceiling) · admin limits (the eleven categories, the periods, the default sample)
• Linked API MCP, available tools (the 59 tools counted by group) · installation (the hosted connector URL, the npx server, the clients) · source on GitHub · Agent Skills (two shipped, three marked coming soon)
• gtm-api.com, product · pricing (our own numbers)
• Related: Best LinkedIn API · LinkedIn MCP server · LinkedIn API pricing · LinkedIn connection limits · LinkedIn integrations

Michael SalamatovGTM engineerRuns the gtm-api.com blog. Covers LinkedIn limits, account health and the API side of outbound.in

Last updated: September 2026 · Every Linked API figure, quotation and count on this page was read off the vendor’s own site, developer documentation, MCP reference and llms.txt on 5 September 2026 and is linked in Sources. Vendors move pricing and claims without notice, so verify before you buy.