Use the AI agent you prefer
Claude, Codex, and other compatible agents can use the same SalesTouch connection. You are not locked into one AI client.
Safe · reliable · scalable
The LinkedIn MCP built for production.
Direct LinkedIn automation from Claude is fragile. SalesTouch gives your AI agent a controlled LinkedIn MCP with secure cloud execution, human-paced queues, tool permissions, and complete logs.
Claude
Codex
SalesTouch
LinkedIn MCP
Securely connected
Find prospects
Read profiles and posts
Send messages
Follow up
Publish content
Connected in real B2B workflows
Under the hood
Follow one request from your agent to LinkedIn.
Six checkpoints run between the moment your agent asks and the moment LinkedIn sees an action. Here is a single invitation, step by step.
Find the best commenters on my last post and invite Camille.
01 / 06
Your agent calls a tool
Claude picks the right SalesTouch tool and sends typed parameters. No browser, no screen scraping, no raw credentials.
- Tool
- Send an invitation
- Profile
- Camille Laurent
- Note
- Loved your post on outbound.
Why the protocol matters
SalesTouch is a LinkedIn MCP.
MCP is an open standard for connecting AI agents to tools. For you, that means SalesTouch works with the agent you already use, exposes only the LinkedIn actions you choose, and returns a clear result every time without giving the AI your browser or your credentials.
The advantage is simple: your AI becomes capable without becoming unrestricted.
What MCP changes for youOpen standard
Use tools, not fragile browser clicks
Your agent asks for a LinkedIn action by name. It does not need to find buttons or control your screen.
Control access action by action
Search, read, message, and publish are separate tools, so you decide exactly what the agent is allowed to use.
Get a clear result every time
Each request comes back with a visible outcome your agent can understand and SalesTouch can log.
Permissions
You decide how much your agent can do alone.
Permissions are set per tool. Start cautious, then open up the routine work once you trust the results.
Read-only
Your agent researches and prepares. Nothing is sent from your account.
- Search and read Allowed
- Invitations Blocked
- Messages Blocked
- Posts and comments Blocked
Approval required
Your agent drafts every action and waits for your go before it reaches LinkedIn.
- Search and read Allowed
- Invitations Approval
- Messages Approval
- Posts and comments Approval
Autonomous within limits
Routine work runs on its own, inside your daily limits and working hours.
- Search and read Allowed
- Invitations Allowed
- Messages Allowed
- Posts and comments Approval
Three common setups. Every tool can be allowed, held for approval, or blocked individually.
When things go wrong
No retry storms. No silent failures.
Every edge case has a defined behavior, and your agent always gets a state it can read and act on.
IfSalesTouchYour agent sees
- LinkedIn asks to slow downThe account pauses and the remaining work is rescheduled for the next eligible slot.Paused
- The daily limit is reachedNew actions stop for that account and the rest moves to tomorrow, in order.Queued for tomorrow
- The LinkedIn session expiresNew work stops, the queue is kept intact, and you are asked to reconnect.Needs reconnection
- Your agent calls a blocked toolThe call is refused before anything happens, and your agent is told why.Refused
- An action needs your reviewIt waits in the queue until you approve or cancel it.Awaiting approval
- A step failsThe error is logged with its details so you or your agent can retry safely.Failed, logged
Execution
Every action has a state. Every state is logged.
Your agent never has to guess whether something happened. It can check the state of any action, and so can you.
- Runs in the cloud, even with your laptop closed.
- State is persisted, so restarts never lose work.
- Queued actions can be rescheduled or cancelled at any time.
- Logs keep the tool, payload, account, duration, and errors for 90 days.
Action lifecycle
RequestedCheckedQueuedRunningDone
Can also become:PausedAwaiting approvalCancelledFailed
Activity log
Last 90 days- 10:42Send an invitationSent
- 10:38Read a profileDone
- 10:31Send a messageAwaiting approval
- 10:24Extract post commentersDone
- 09:58Send an invitationQueued
Multiple accounts
One endpoint. Every account in its own lane.
Your agent reaches every connected account through the same MCP, while each LinkedIn identity keeps its own location, session, limits, and queue.
| Kept separate |  Thomas MartinFounder |  Sarah LopezHead of Sales |  Marcus JohnsonClient account |
|---|---|---|---|
| Login location | Paris, France | New York, US | Berlin, Germany |
| LinkedIn session | Own session | Own session | Own session |
| Invitations per day | 30 | 25 | 20 |
| Working hours | 09:00 to 18:00 | 08:00 to 17:00 | 10:00 to 19:00 |
| Queue | 12 queued | 4 queued | 7 queued |
| Activity logs | Separate | Separate | Separate |
A limit reached or a pause on one account never slows down the others.
The infrastructure behind every action
Sending one LinkedIn action is easy. Running it safely every day is the hard part.
Build it yourself and you are not building one integration. You are taking responsibility for the connection, safety rules, execution engine, logs, and every account you add. SalesTouch ships that entire operating layer by default.
Your team owns the stack
Build it yourself
Before your agent can reliably send one message, you need five production systems working together.
Secure LinkedIn connection
Source residential IPs, persist sessions, encrypt authentication, detect disconnects, and reconnect safely.
Account safety engine
Create daily limits, working hours, human pacing, cooldowns, and rules for every risk signal.
Durable execution engine
Run queues, scheduling, retries, deduplication, and persistent state that survives restarts.
Observability and recovery
Trace every request, status, error, and retry, then build ways to pause, cancel, and recover safely.
Multi-account isolation
Separate credentials, connections, queues, limits, and state for every LinkedIn identity you add.
And it never becomes “done.”
Monitor it every day, patch it when LinkedIn changes, repair broken sessions, and own every failure across every account.
The stack is already running
Connect SalesTouch
One MCP connection gives your agent a production-ready LinkedIn operating layer.
Secure LinkedIn connection
Residential IP, encrypted authentication, session health, and reconnection are managed per account.
Account safety engine
Limits, pacing, working hours, and cooldowns are checked before every action.
Durable execution engine
Cloud queues, scheduling, retries, and persistent state keep every workflow running.
Observability and recovery
Explicit statuses and 90-day activity logs make every action inspectable and recoverable.
Multi-account isolation
Each account stays isolated while your agents control all of them through the same MCP.
You design the workflow. SalesTouch keeps LinkedIn execution safe, reliable, and scalable.
FAQ
LinkedIn MCP questions, answered.
Everything you need to know before connecting your AI agent to LinkedIn.
A LinkedIn MCP is a connector that exposes LinkedIn capabilities as typed tools an AI agent can call through the Model Context Protocol. SalesTouch adds OAuth permissions, account protection, queues, cloud execution, and logs around those tools so Claude, Codex, ChatGPT, and other MCP clients can use LinkedIn through one controlled endpoint.
Get started
Your LinkedIn MCP is 10 minutes away.
Connect SalesTouch to the AI agent you already use and give it reliable LinkedIn tools in minutes.
7 days free · No browser extension · Cancel anytime