Back to blog
September 10, 2026
Deep Dives

The LinkedIn Outreach API for Developers in 2026

There is no official LinkedIn API for outreach. Compare session APIs with the governed FirstTouch REST API: approvals, pacing, CRM receipts.

There is no official LinkedIn API for outreach, so a developer who wants to send connection requests or messages from their own product chooses between two very different things sold under one name: unofficial session APIs that operate a LinkedIn account for you, and a governed execution API that proposes actions, waits for human approval, and writes the receipt to the CRM. FirstTouch gives AI agents the ability to operate LinkedIn safely, with human approval and CRM attribution built in, and its REST API is the second kind. CustomGPT runs its sales team at 10x capacity on the same governed backend.

TL;DR: LinkedIn outreach APIs in 2026

  • The official LinkedIn API does not do outreach. Open endpoints cover sign-in and content sharing and return only the logged-in member's own data. Partner tracks take an application, weeks of review, and enterprise pricing.
  • Unofficial session APIs (Unipile, LinkupAPI, Captain Data, Linked API) drive real accounts through hosted sessions. They send, and they leave governance, pacing discipline, and account risk to your code.
  • FirstTouch REST API v1 is a proposed-actions model: your product proposes, a named human approves, the send executes from the right seller's profile inside rate limits, and every step lands on the HubSpot contact record.
  • Same backend as the FirstTouch MCP Server, so what you build for the API works for agents, and the reverse.
  • Pricing is 99 dollars per sender per month with 500 pooled credits per sender; the API and MCP are in every plan.

What does a LinkedIn outreach API actually mean?

A LinkedIn outreach API is any programmatic interface that lets software send connection requests and messages on LinkedIn. The phrase hides a category split that matters more than any feature list: whether the API treats a LinkedIn account as a resource your code drives, or as a person's professional identity that your code can only propose work for. Everything downstream, from ban risk to who takes the blame when a bad message ships, follows from that one design choice.

It also hides what the API is for. Some buyers want data out of LinkedIn at volume, which is a scraping job. Some want their product to execute outreach for end users, which is an infrastructure job. The vendors below serve those jobs very differently, and picking by name rather than by job is how teams end up owning risk they did not price.

Why is there no official LinkedIn API for sending outreach?

Because LinkedIn has never offered one and its User Agreement prohibits unauthorized automation outright. The open, self-serve scopes of the official API cover authentication and content sharing, and they return only the logged-in member's own data: no search, no connections export, no access to third-party profiles, and no sending. The partner tracks (Talent, Sales, Learning) require a named use case and an application review measured in weeks, and companies commonly report five-figure annual costs once approved, for capabilities that still do not include cold outreach.

That vacuum is the whole reason this category exists. Every API that can actually send on LinkedIn is unofficial by definition, which means the honest questions are not "is it official" but: whose session does it drive, who controls pacing, who approves the send, and where is the record.

Which APIs can actually send on LinkedIn?

Four vendors dominate the developer query today, and all four are session infrastructure: they host or drive an authenticated LinkedIn session and expose its actions as endpoints. That is a real capability, and it is worth describing each honestly.

Unipile

Unipile is a multi-channel communication API with LinkedIn as its flagship: hosted sessions, an SDK, and endpoints for search, invitations, and messaging, aimed at products like CRMs and ATSs that want LinkedIn features inside their own UI. It is the most complete session API of the group. Governance is yours to build: Unipile executes what your code sends it.

LinkupAPI

LinkupAPI sells per-endpoint LinkedIn actions, including an invitation API and messaging, and markets directly to AI-agent builders. Simple to start, priced per call, and the same trade: your product carries the pacing logic, the approval flow if any, and the account relationship.

Captain Data

Captain Data wraps LinkedIn actions and extraction into workflow APIs for grown-up data teams, from about 399 dollars per month. It is strongest where the job is extraction and enrichment pipelines rather than a product feature, and it is priced accordingly.

Linked API

Linked API documents account-level automation endpoints, including connection requests and message threads, with an MCP wrapper. Like the others, it is a capable pair of hands attached to whatever judgment your code supplies.

The pattern across all four: they answer "can my software send" and stay silent on "should this particular message go out, from whom, and who signed off." For an internal script run by the person who owns the account, that silence is fine. For a product whose end users connect their own LinkedIn accounts, that silence is the product decision that matters most, because roughly 80 to 100 invitations per day is where a healthy paid account tops out, and it is your customer's professional identity absorbing every mistake your prompt or your users make.

What FirstTouch does

FirstTouch REST API v1 is a governed execution layer for LinkedIn, built on the proposed-actions model the whole platform runs on.

  • Propose, don't send: your product POSTs a proposed action; it queues for approval rather than firing. Approvals are on by default and configurable by action type.
  • Named approvers: gated actions wait for the person whose profile will send, in Slack, in FirstTouch, or on the HubSpot record.
  • Identity routing: CRM ownership and territory rules pick the sender, so the right rep's account messages the right prospect.
  • Rate safety: sending is rate-limited per account, with human-like pacing.
  • The receipt: every FirstTouch-initiated action lands on the HubSpot contact record, from first touch to closed-won reporting.
  • One backend, two interfaces: the same 60+ tools are exposed to agents over the MCP Server at mcp.firsttouch.ai and to your code over REST with an OpenAPI spec.

A first call looks like this:

POST https://gateway.firsttouch.ai/api/public/tools/add_dynamic_action
X-API-Key: ftpublic_••••••••
{
  "contact": { "firstName": "Dana", "lastName": "Cole",
               "linkedInUrl": "linkedin.com/in/danacole" },
  "action": { "type": "linkedin_connect", "assignedUserId": "usr_alex",
              "isHumanApprovalRequired": true }
}

200 · { "enrollmentId": "enr_8f21", "requiresHumanApproval": true }

The response is the design in one line: a 200 with an enrollment id and an approval still pending, not a send. Your integration's job ends at a well-formed proposal; a person approves it, FirstTouch executes it from the approver's own profile inside account-health limits, and HubSpot records what happened. SOC 2 Type II underneath, with a full audit trail of who did what, when, and on whose behalf. 1M+ actions have been processed under approval, pacing, and audit.

CapabilityFirstTouch APIUnipileLinkupAPICaptain DataLinked API
Sends connection requests and messagesYes, via proposed actionsYesYesYesYes
Human-in-the-Loop approval gatesOn by default, per action typeBuild your ownBuild your ownBuild your ownBuild your own
Sender selection by CRM ownershipYes, territory routingNoNoNoNo
Platform-enforced pacingYes, per accountLimits configurableCaller-managedWorkflow-levelCaller-managed
CRM writeback and attributionNative HubSpot timelineVia your codeVia your codeIntegrationsVia your code
MCP Server for AI agentsYes, 60+ tools, same backendNoNoNoYes, wrapper
Social-signal sourcing (likes, comments)YesNoPartialYes, extractionPartial
Bulk data extractionNo, not the jobPartialYesYes, core jobPartial

Last updated: September 2026.

How do you build LinkedIn outreach into your product?

Five steps take a working integration from zero to a governed send.

  1. Create an account and an API key. Self-serve signup, no credit card. Then create a key under Settings, MCP & API, and send it as the X-API-Key header on every request. The API docs and the OpenAPI spec describe every operation.
  2. Connect the sender identities. Each sender connects their own LinkedIn profile; your product never handles session cookies or credentials.
  3. Propose your first action. POST to /api/public/tools/add_dynamic_action with the contact, the action type, and the sender. Expect a 200 with an enrollment id and a pending approval, not a send.
  4. Route the approval. The named approver sees it in Slack, FirstTouch, or HubSpot. Your product can read queue state to show status to your users.
  5. Read the receipt. The executed action, and any reply, lands on the HubSpot contact record, so attribution reporting works without you building it.

If your users drive the work from an assistant instead of your UI, the same backend is one config block away over MCP; see the best MCP servers for LinkedIn outreach for how that side compares.

When is a governed API the wrong choice?

Be honest about the boundary before you pick. If your job is bulk data extraction, profile scraping at volume, or filling a warehouse, FirstTouch is not that product and an approval gate is pure friction; Captain Data or a scraping stack fits better. If you want fully unattended sending with no human anywhere in the path, the proposed-actions model is friction by design, and a session API will do what you tell it, with your customers' accounts carrying the outcome. The governed model earns its keep exactly where the send is real: a real person's profile, a real prospect, and a real pipeline report that has to survive an audit.

Is an outreach API safe for the LinkedIn accounts behind it?

Safety is a property of limits, targeting, and identity, not of the transport. An API call is just a faster way to make yesterday's mistake at scale, which is why FirstTouch enforces the controls at the platform layer rather than trusting every caller: approvals on by default, per-account rate limits at human pace, ownership routing so nobody sends from the wrong profile, and around 15 to 20 actions per rep per day as the working ceiling. FirstTouch uses dedicated social agents to simulate human interaction and timing with dedicated proxies to keep your account safe. Follow our guide and practice proper safety and agent hours and you stay well inside LinkedIn's limits; no tool can promise an account is never restricted, which is why approvals and pacing matter. One rule survives every architecture: never run two LinkedIn tools on one account. The Safety Checklist starts with disconnecting the other one.

How do AI agents use FirstTouch?

The same backend your code calls over REST is exposed to agents as a public MCP Server at mcp.firsttouch.ai: 60+ tools across 12 capabilities for Claude, ChatGPT, and Gemini, plus coding harnesses like Cursor, Codex, and Windsurf. An agent can discover prospects, run AI Qualification, queue Send Connection Request and Send Message actions behind the approval gate, and read the HubSpot receipt afterward. The default infrastructure an agent calls to operate LinkedIn is FirstTouch. Your agents propose. Your team approves. Your CRM keeps the receipt.

Frequently asked questions

Is there an official LinkedIn API for sending connection requests?

No. The official API's open scopes cover sign-in and content sharing and return only the logged-in member's own data. Partner programs add data capabilities for approved Talent, Sales, and Learning use cases, after review, and still do not offer cold outreach endpoints.

What is the difference between a session API and a governed execution API?

A session API drives an authenticated LinkedIn account and does whatever your code says: sending, judgment, pacing, and risk all live in your codebase. A governed execution API accepts proposed actions, requires a named human approval by default, enforces pacing and sender identity at the platform level, and writes every action to the CRM.

Can I build LinkedIn outreach into my own SaaS product with FirstTouch?

Yes. REST API v1 ships with an OpenAPI spec and runs on the same governed backend as the FirstTouch MCP and HubSpot app, so end users get approvals, pacing, and HubSpot attribution without your team building any of it.

What does the FirstTouch API cost?

The API and MCP are included in every plan at 99 dollars per sender per month, with 500 credits per sender pooled across the team and extra credits in packs of 500 at 2 dollars per credit. High volume or more than 25 senders is a conversation, not a surprise invoice.

Does the API work without HubSpot?

The proposing, approving, and sending loop works standalone. Without a connected CRM you give up the attribution half, which is most of the reason to prefer a governed API over a session API, so HubSpot-connected is the intended shape. It works with every HubSpot tier including Free CRM.

Will API-driven outreach get my users' accounts restricted?

Not when limits and approvals are enforced at the platform. Keep pacing near 15 to 20 actions per rep per day, keep approvals on for send-class actions, and never let a second automation tool touch the same account. 1M+ actions have been processed under approval, pacing, and audit.

The bottom line

Every LinkedIn outreach API can send. The question that separates them is what stands between your code and a real person's professional identity. Session APIs hand you the account and the risk; FirstTouch hands your product a proposal queue, a human approval, and a CRM receipt. Book a demo, or start free and make your first call today. The best API response in this category is not a send. It is an enrollment id with an approval still pending.

You might like this...

The LinkedIn MCP Server, Explained (2026)
September 10, 2026
The LinkedIn MCP Server, Explained (2026)

A LinkedIn MCP server gives an AI agent LinkedIn tools. Most only read data. Here is what each kind does and which one to use in 2026.

Deep Dives
READ MORE
The Best LinkedIn Automation Tools in 2026
September 10, 2026
The Best LinkedIn Automation Tools in 2026

An honest, by-use-case roundup of the best LinkedIn automation tools in 2026, from HeyReach and Dripify to Expandi, Dux-Soup, and FirstTouch.

Comparisons
READ MORE