Best LinkedIn MCP Server 2026: 8 Tools Compared

Published On October 30, 2026

Quick answer: Kondo is the best LinkedIn MCP server for inbox and DM management. It runs through the user's own browser session, never shares credentials, paces activity automatically, and is draft-only — meaning the AI cannot send anything without human review. Where send capability is required and the trade-offs of a hosted session are accepted, LinkMCP is the strongest runner-up.

7 LinkedIn MCP servers compared; LinkedIn ships no first-party MCP server, so every option is third-party.
Connection method is the deciding factor for account safety: local browser sessions are the safer path, hosted sessions enable sending but carry account-restriction risk, and the official API cannot access personal DMs.
Match the tool to the job: inbox triage, confirm-first outbound, developer integrations, and free experimentation each call for a different option.
For account-safe inbox and DM management, Kondo's LinkedIn MCP server runs through the user's own browser session, paces activity automatically, and stays draft-only from $28/mo.

TL;DR: 7 LinkedIn MCP Servers Compared

What a LinkedIn MCP Server Actually Is

A LinkedIn MCP server is the bridge between an AI assistant — Claude, Cursor, ChatGPT, Gemini, Copilot — and a LinkedIn account. The Model Context Protocol defines a standard way for an AI to call tools: read the inbox, search profiles, draft a reply, apply a label. Without an MCP server, the AI has no way to reach LinkedIn at all.

This article is not about LinkedIn CRMs, outbound sequencing platforms, or content scheduling tools. It is specifically about MCP servers: software that gives a conversational AI direct, structured access to LinkedIn.

One fact anchors everything that follows: LinkedIn ships no first-party MCP server. Every option in this list is built by a third party. That matters for both safety and expectation-setting.

Connection Method: The Difference That Decides Everything

Before comparing features or prices, the connection architecture must be established. This single decision determines account risk, what the AI can actually do, and where data resides.

There are four methods:

1. The user's own browser session. The tool runs locally and acts through the currently active, logged-in browser. There is no third-party session host and no credentials handed over. This is the safer path. The trade-off is that actions are typically limited to reading and drafting — send capability is rare.

2. Hosted managed session. The user provides a session cookie or credentials to a third-party vendor, which runs a browser on its servers acting as the account holder. This enables 24/7 automation and send capability, but it operates outside LinkedIn's approved developer platform, with account restriction risk. The account holder remains responsible for everything that happens. The vendor acts as the account holder — this is a liability statement rather than a marketing claim.

3. Official Partner API. The tool uses LinkedIn's documented APIs via OAuth. This is terms-compliant, but it comes with a firm ceiling: the official LinkedIn API does not expose personal DM inboxes. Microsoft confirms no public API supports general-purpose access to personal messages. Where inbox triage is the objective, the official API cannot deliver it.

4. Scraper / unofficial API. Browser automation (Selenium, Playwright) or reverse-engineered private endpoints. Free and technically capable, but it operates outside LinkedIn's approved developer platform, breaks whenever LinkedIn updates its site, and carries the highest account-restriction risk of the four methods. A single poorly paced loop flags an account as automated faster than most operators expect.

How We Evaluated

Each tool is scored across four criteria:

Connection method and account safety (35%): The highest weight because account loss is the irreversible outcome. The assessment covers whether the method is session-based (local vs. hosted) or API-based, and what LinkedIn's User Agreement and Professional Community Policies state about it.
Inbox capability (30%): The ability to read conversations, find unread and awaiting-reply threads, search message history, draft replies, apply labels, and archive. This is the core use case for most AI-assisted workflows.
Action depth (20%): People search, profile lookup, company data, invitation management, post interactions — capability beyond the inbox.
Price and setup (15%): Trial availability, monthly cost, and deployment friction. Setup friction is a real barrier: cookie authentication errors and unclear install instructions are among the most common issues with self-hosted options.

The 7 Best LinkedIn MCP Servers in 2026

1. Kondo — Best LinkedIn MCP Server for Inbox and DM Management

Verdict: Among the options here, Kondo has one of the lowest account-risk profiles on this list. It uses the user's own browser session, keeps credentials on the local machine, and has no send tool on any platform. For inbox triage, it covers more ground than any other tool on this list.

Kondo's LinkedIn MCP server connects via a single MCP endpoint (https://relay.trykondo.com/mcp) and integrates with Claude, Cursor, and other MCP-compatible assistants. It authenticates through the active browser session — no cookie is extracted or transmitted to a third party. Messages and contacts stay on the user's computer, and message content goes only to the AI provider the user connects.

The inbox toolset is the deepest available: list and search conversations, surface unread and awaiting-reply threads, read full message history, draft responses, apply and remove labels, add notes, set follow-up reminders, and archive. Activity is paced and backs off automatically when LinkedIn is busy, which reduces the signature of automated behaviour.

The limitation is deliberate and is the design position. Kondo has no send tool. Every draft it produces requires human review and manual send. For users who need an AI to automate outreach and send autonomously, it does not fit. For users who need an AI to handle the cognitive load of inbox management without granting a third party a separate LinkedIn session and a copy of their message history, it is purpose-built for exactly that.

Disconnect at any time via Kondo > Settings > MCP. For the full detail on how Kondo handles your session and data, read the Kondo safety overview.

Pricing: Starts from $28/mo, 14-day money-back guarantee.

Who should skip this tool: Anyone who needs the AI to send messages or connection requests autonomously. Kondo is deliberately built to prevent that.

2. LinkMCP — Best Hosted-Session Option for Research and Inbox

Verdict: LinkMCP offers the widest tool coverage of any hosted linkedin mcp server — 33 tools spanning people search, Sales Navigator, profiles, companies, jobs, posts, invitations, and conversations. It includes 8 write tools, one of which is send. Where that capability is required and the trade-offs are accepted, this is the strongest hosted option.

The connection method is a hosted managed session. LinkMCP is transparent that it is not an official LinkedIn partner API. The session runs on vendor infrastructure. Built-in rate limiting, warm-up limits for new accounts, and hard monthly caps reduce exposure, but they do not eliminate it. The account holder remains responsible under LinkedIn's terms.

Silent data drift is a known risk with any API wrapper of this kind. When LinkedIn changes a profile field or search response shape, downstream tools can return partial results without surfacing an error. This is a known fragility across the hosted-session category.

Pricing: Free 7-day trial. Starter $19/mo, Pro $49/mo, Max $199/mo. Top-ups at $10.

Who should skip this tool: Users not comfortable sharing a LinkedIn session with a third-party vendor, or those whose account cannot absorb the terms-of-service risk.

3. Unipile — Best for Developers

Verdict: Unipile is not an inbox tool for individuals. It is a messaging API that aggregates LinkedIn, WhatsApp, Instagram, and email into a single interface and exposes that via MCP. For teams building a system that must sync and route messages across channels, Unipile is the right infrastructure piece.

Read and send actions work through the API — programmatic sending from code, not scheduled automation. Data is synced to and resides on Unipile's cloud, a straightforward trade-off for a managed API service. For personal inbox triage through an AI assistant, the architecture does not fit: Unipile is infrastructure for developers building messaging features; Kondo is an inbox professionals use every day.

Pricing: Approximately $55/mo.

Who should skip this tool: Non-developers, and anyone seeking a ready-to-use tool to connect Claude or Cursor to a personal LinkedIn inbox. Unipile requires API integration and is not designed for human-first inbox management.

4. Fintalio — Best Confirm-First Outbound Prospecting MCP

Verdict: Fintalio's standout feature is confirm-first execution. The AI can plan and queue actions — connection requests, follow-up messages, sequence steps — but a human must approve each one before it fires. For outbound sales workflows where full automation is too risky and full manual effort is too slow, this middle ground is operationally useful.

The tool set covers 18 actions: 9 read, 5 manage, and execution tools gated behind the confirm step. Sequences and contacts sit at the centre of its design. Inbox triage is peripheral. Fintalio uses scoped tokens for access controls, and the connection method is a hosted managed session — so the same account-restriction considerations apply as with LinkMCP.

Pricing: Flat €69/mo, 120 requests per minute per token.

Who should skip this tool: Users whose primary need is reactive inbox management. Fintalio's architecture is built around structured outbound campaigns, not triage of inbound messages.

5. Composio — Best for Agent Workflow Glue

Verdict: Composio connects LinkedIn to a library of 1,500+ other tools, making it the right choice when LinkedIn is one node in a larger automated workflow rather than the primary surface under management.

Its LinkedIn integration uses the official Partner API path, which keeps it inside LinkedIn's approved developer platform but limits it. Personal DM inbox access is not available through this method. The focus lands on company page management and post interactions — appropriate for brand and content workflows, not personal outreach or inbox triage.

The Tool Router feature intelligently selects the right tool for a given agent task. The hobby plan includes 100,000 free tool calls, which covers substantial experimentation.

Pricing: Hobby plan free (100k tool calls). Paid tiers available.

Who should skip this tool: Anyone focused on personal inbox management, DM triage, or people-level prospecting. Composio's LinkedIn coverage is thin on those specific features by design.

6. Open-Source Scrapers (e.g., stickerdaniel/linkedin-mcp-server) — Best Free Option, Highest Risk

Verdict: The stickerdaniel/linkedin-mcp-server project (Apache 2.0) is the most capable free option and the highest-risk one. It runs browser automation on the local machine, ships approximately 20 tools including read of conversations and management of connection requests, and requires Docker to run.

Setup is demanding. Cookie authentication errors are common, CAPTCHA handling rests with the operator, and the tooling breaks when LinkedIn updates its frontend. These are not edge cases — they are the standard cost of maintaining a scraper-based system. The setup experience here is among the most difficult on this list, and the friction is structural rather than a documentation gap.

LinkedIn's User Agreement restricts automated access to the site. Detection risk here is higher than with commercial tools that pace requests through tested rate-limiting logic. One aggressive loop produces an account signal that reads as automated within minutes.

A practical note: several open-source projects in this category are sponsored by or affiliated with hosted services that offer to "handle the auth and infrastructure." That offer moves the user into the hosted-session model and its associated risks, rather than solving the problem.

Who should skip this tool: Anyone not comfortable with Docker, CAPTCHA handling, and command-line configuration — and anyone using a primary LinkedIn account they cannot afford to lose.

7. adhikasp/mcp-linkedin — Lightweight Feed and Job Search Only

Verdict: This is an experimental, read-only open-source server scoped to feed reading and job search. It uses direct email and password authentication, which hands credentials to the tool — an architectural tradeoff on top of the standard scraper risks.

The project is explicitly marked as experimental and not intended for production use. The tool set is narrow. There is no inbox access and no write capability.

Who should skip this tool: Almost everyone. This exists for developers testing a narrow, specific integration on accounts that are not primary. It is not a practical tool for professional use.

Conclusion: Which LinkedIn MCP Server Fits Your Situation?

Inbox management, maximum account safety: Use Kondo. The own-session, draft-only model is purpose-built for this and carries a lower account-restriction risk than the hosted-session and scraper options on this list.

Send capability with the hosted-session trade-off accepted: Use LinkMCP. It has the deepest toolset in the hosted category, transparent rate limiting, and a free trial to test before committing.

A custom messaging integration across platforms: Use Unipile. It is an API-first infrastructure tool, not a personal inbox product — which is exactly right for teams engineering a system.

Outbound sequences with a human approval step: Use Fintalio. The confirm-first model is a practical middle ground between full automation and fully manual outreach.

LinkedIn connected to a broader agent workflow: Use Composio. Its 1,500+ tool library makes LinkedIn one piece of a larger automated system, and the free hobby tier keeps experimentation costs low.

Free experimentation on a non-critical account: Use an open-source scraper like stickerdaniel/linkedin-mcp-server, with full awareness that setup is complex, it breaks on LinkedIn updates, and ban risk is real.

Further reading: The LinkedIn inbox management tool guide, the best LinkedIn DM management tools, how to set reminders for LinkedIn messages, combining Sales Navigator and your LinkedIn inbox, and shortcuts for the LinkedIn inbox.

Frequently Asked Questions

Is there an official LinkedIn MCP server?

No. LinkedIn does not offer a first-party MCP server; every option is third-party or community-built. LinkedIn's developer catalog lists REST APIs only. Treat any vendor claim of official MCP status or partnership as a prompt to verify the connection method and terms compliance.

Will using a LinkedIn MCP server get me banned?

It depends on how it connects. Hosted sessions and scrapers operate outside LinkedIn's approved developer platform and carry account-restriction risk; local browser-session tools for read and draft actions carry lower risk. No third-party tool can guarantee zero risk, so automation should be avoided on accounts that cannot be lost.

Can a LinkedIn MCP server send messages for me?

Some can. Hosted-session and scraper tools can send DMs and connection requests; official API tools cannot send personal DMs. Draft-only tools like Kondo let the AI draft replies but require human review and manual send, which keeps a human in control.

What is the difference between an inbox tool and a prospecting tool?

Inbox tools handle reactive work: unread messages, thread summaries, labels, reminders, and reply drafts. Prospecting tools handle proactive outreach: lead lists, sequences, and connection requests. Kondo and LinkMCP focus on the inbox; Fintalio focuses on confirm-first outbound sequences.

Can a LinkedIn MCP server read my personal DM inbox?

Only if it uses a session-based connection, either the user's own local browser session or a hosted one. Tools built on the official LinkedIn API cannot access personal DMs. Where a vendor claims official-API inbox access, proof should be requested and the connection method scrutinized.

How do I connect a LinkedIn MCP server to Claude or Cursor?

It varies by tool. Kondo's LinkedIn MCP server uses one MCP endpoint in the assistant's settings, with no cookie extraction. Open-source tools often require Docker and a Claude Desktop config file. Hosted and OAuth tools usually require a sign-in or a session cookie, which is where most setup errors occur.