
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.
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.
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.
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.
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 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 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 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 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.
FirstTouch REST API v1 is a governed execution layer for LinkedIn, built on the proposed-actions model the whole platform runs on.
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.
| Capability | FirstTouch API | Unipile | LinkupAPI | Captain Data | Linked API |
|---|---|---|---|---|---|
| Sends connection requests and messages | Yes, via proposed actions | Yes | Yes | Yes | Yes |
| Human-in-the-Loop approval gates | On by default, per action type | Build your own | Build your own | Build your own | Build your own |
| Sender selection by CRM ownership | Yes, territory routing | No | No | No | No |
| Platform-enforced pacing | Yes, per account | Limits configurable | Caller-managed | Workflow-level | Caller-managed |
| CRM writeback and attribution | Native HubSpot timeline | Via your code | Via your code | Integrations | Via your code |
| MCP Server for AI agents | Yes, 60+ tools, same backend | No | No | No | Yes, wrapper |
| Social-signal sourcing (likes, comments) | Yes | No | Partial | Yes, extraction | Partial |
| Bulk data extraction | No, not the job | Partial | Yes | Yes, core job | Partial |
Last updated: September 2026.
Five steps take a working integration from zero to a governed send.
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.
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.
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.
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.
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.
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.
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.
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.
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.
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.
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.

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.

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

Build an AI agent for LinkedIn outreach: point it at an MCP server that finds people, queues sends for your approval, and logs each touch to your CRM.