LinkedIn MCP Server

Disclaimer: LinkedIn is a registered trademark of LinkedIn Corporation and its affiliates. LinkedIn MCP Server is an independent, unofficial project and is not affiliated with, endorsed by, sponsored by, or associated with LinkedIn Corporation or its affiliates.

A LinkedIn MCP server to find jobs, search people, research companies, manage your network, publish and engage with posts, and read or send messages.

Features

Actions that change LinkedIn use an immutable prepare-and-execute flow and request confirmation by default. Every capability is task-specific; the server does not expose unrestricted browser, click, navigation, JavaScript, or network access.

See the capability matrix for exact filters, supported formats, inputs, outputs, limits, and unsupported features.

Related MCP server: LinkedIn MCP Pro Max

Installation

Or add this to .vscode/mcp.json:

{
  "servers": {
    "linkedin-mcp": {
      "command": "uvx",
      "args": ["--from", "linkedin-mcp-local", "linkedin-mcp", "serve", "--transport", "stdio"]
    }
  }
}

See the VS Code MCP documentation.

Add this to ~/.codex/config.toml:

[mcp_servers."linkedin-mcp"]
command = "uvx"
args = ["--from", "linkedin-mcp-local", "linkedin-mcp", "serve", "--transport", "stdio"]
startup_timeout_sec = 60
tool_timeout_sec = 900
default_tools_approval_mode = "auto"

Codex CLI, the Codex IDE extension, and ChatGPT Desktop share this local configuration. Restart the client after saving it. See the Codex MCP documentation.

claude mcp add --scope user --transport stdio linkedin-mcp -- \
  uvx --from linkedin-mcp-local linkedin-mcp serve --transport stdio

Check it with claude mcp list. See the Claude Code MCP documentation.

Download the .mcpb file, then open Settings → Extensions → Advanced settings → Install Extension. See Claude Desktop's extension documentation.

For manual setup, open Cursor Settings → MCP → Add new MCP server, set the command to uvx, and set the arguments to --from linkedin-mcp-local linkedin-mcp serve --transport stdio. See the Cursor MCP documentation.

In ~/.gemini/settings.json, add a local server named linkedin-mcp under mcpServers with command uvx and arguments --from linkedin-mcp-local linkedin-mcp serve --transport stdio, then run gemini mcp list. See the Gemini CLI MCP documentation.

Open Settings → AI → Manage MCP Servers, or add a local server named linkedin-mcp to ~/.codeium/windsurf/mcp_config.json with command uvx and arguments --from linkedin-mcp-local linkedin-mcp serve --transport stdio. See the Windsurf MCP documentation.

Open MCP Servers → Configure, then add a local server named linkedin-mcp with command uvx and arguments --from linkedin-mcp-local linkedin-mcp serve --transport stdio. See the Cline MCP documentation.

Open Roo Code's MCP settings and add a local server named linkedin-mcp to the global file or .roo/mcp.json, using command uvx and arguments --from linkedin-mcp-local linkedin-mcp serve --transport stdio. See the Roo Code MCP documentation.

Or add a local server named linkedin-mcp to ~/.kiro/settings/mcp.json with command uvx and arguments --from linkedin-mcp-local linkedin-mcp serve --transport stdio. See the Kiro MCP documentation.

Add this to Zed settings:

{
  "context_servers": {
    "linkedin-mcp": {
      "command": "uvx",
      "args": ["--from", "linkedin-mcp-local", "linkedin-mcp", "serve", "--transport", "stdio"]
    }
  }
}

See the Zed MCP documentation.

Or open Program → Install → Edit mcp.json and add a local server named linkedin-mcp with command uvx and arguments --from linkedin-mcp-local linkedin-mcp serve --transport stdio. See the LM Studio MCP documentation.

Or add a custom stdio extension with command uvx and arguments --from linkedin-mcp-local linkedin-mcp serve --transport stdio, or start one CLI session with:

goose session --with-extension \
  "uvx --from linkedin-mcp-local linkedin-mcp serve --transport stdio"

See the Goose documentation.

For OpenCode v2, add this to opencode.json:

{
  "$schema": "https://opencode.ai/config.json",
  "mcp": {
    "servers": {
      "linkedin-mcp": {
        "type": "local",
        "command": ["uvx", "--from", "linkedin-mcp-local", "linkedin-mcp", "serve", "--transport", "stdio"]
      }
    }
  }
}

See the OpenCode MCP documentation.

Open Settings → Tools → AI Assistant → Model Context Protocol (MCP), click Add, set the command to uvx, and set the arguments to --from linkedin-mcp-local linkedin-mcp serve --transport stdio. See the JetBrains MCP documentation.

Create .continue/mcpServers/linkedin-mcp.json with a local server named linkedin-mcp, command uvx, and arguments --from linkedin-mcp-local linkedin-mcp serve --transport stdio. See the Continue MCP documentation.

Open Settings → AI → Manage MCP Servers → Add, name the server linkedin-mcp, set the command to uvx, and set the arguments to --from linkedin-mcp-local linkedin-mcp serve --transport stdio. See the Warp MCP documentation.

Usage

First-time LinkedIn login

After installation, restart your MCP client. The server opens LinkedIn in a browser window where you can sign in and complete MFA or any checkpoint. On this first use it creates its own persistent Chromium profile automatically. Your session is saved in that profile and reused across restarts.

If you have not started the server yet and want to log in manually, create the dedicated profile first:

uvx --from linkedin-mcp-local linkedin-mcp profile create
uvx --from linkedin-mcp-local linkedin-mcp login

The server never asks for or stores your LinkedIn password.

Local session controls

Use these commands without locating PIDs or deleting lock files manually:

# Show or gracefully stop the process that owns this account
uvx --from linkedin-mcp-local linkedin-mcp status
uvx --from linkedin-mcp-local linkedin-mcp stop

# Inspect the dedicated Chromium profile or sign out visibly
uvx --from linkedin-mcp-local linkedin-mcp profile status
uvx --from linkedin-mcp-local linkedin-mcp logout

For a clean profile, run linkedin-mcp profile reset. The command asks for confirmation, archives the old profile, and creates a replacement. See Configuration for the complete lifecycle.

Try it

Ask your MCP client naturally:

Find jobs

Find remote software engineering jobs in India posted this week with Easy Apply.

Research people

Find engineering managers at Stripe in India and show me their relevant experience.

Research companies

Find fintech companies in Bengaluru with 51–200 employees and summarize each company.

Explore posts

Find recent LinkedIn posts about AI agents and summarize the most useful discussions.

Manage your network

Show my latest received connection requests.

Send a connection request to <profile URL> with the note <note>.

Send messages

Send <message> to <profile URL>.

Approval modes

Account-changing execute tools request confirmation by default. Approval is a setting of the MCP client, not something an agent can grant itself. To let a Codex scheduled task publish posts unattended while every other LinkedIn action keeps its default behavior, add this explicit per-tool approval:

[mcp_servers."linkedin-mcp".tools."linkedin.posts.create.execute"]
approval_mode = "approve"

Restart Codex after changing its configuration. The post still goes through the same immutable draft, scope, payload-hash, idempotency, identity, and visible postcondition checks. Avoid approving the entire server when only one action is needed. See Configuration for the full policy model.

Architecture

%%{init: {"flowchart": {"nodeSpacing": 55, "rankSpacing": 65}, "themeVariables": {"fontSize": "20px"}}}%%
flowchart LR
    A["MCP Clients<br/>Codex · Claude · Cursor"] --> B["stdio bridges or<br/>loopback HTTP"]
    B --> C["Shared Local Runtime<br/>Typed LinkedIn Tools"]
    C --> D["Fair Per-Client Queue"]
    D --> E["One Atomic Browser Operation<br/>Fresh Page · Global Pacing"]
    E -->|"Visible UI only"| F["LinkedIn"]
    E <--> G["One Chromium Context<br/>Persistent Profile"]
    C -. "Confirmation previews" .-> A

Everything runs locally. There is no hosted backend, telemetry, database, external queue, LangGraph runtime, or credential service. Browser cookies live only in the local Playwright profile. The first client starts one shared local runtime; later clients attach to it, and fair scheduling gives each client a turn between complete tool calls. Operation state lasts only for that runtime. Read the full architecture and privacy policy.

Configuration

Common settings control:

• enabled LinkedIn surfaces, capability scopes, and effect classes;

• client-side interactive or explicit per-tool approval behavior;

• the persistent browser profile and headed/headless operation;

• the local attachment directory;

• internal pacing, queue capacity, and bounded collection traversal; and

• stdio or loopback-only Streamable HTTP transport.

See Configuration for ready-made permission presets, every environment variable, local HTTP sharing, and the container image.

Privacy Policy

The server has no maintainer-operated backend, analytics, advertising, or telemetry. LinkedIn receives normal visible-UI requests, and the invoking MCP client receives tool results under its own data policy; the project sends nothing to the maintainer. Read the complete privacy policy for processing, storage, sharing, retention, and deletion details.

Safety

Use of this software is at your own risk. You are solely responsible for complying with LinkedIn's User Agreement, applicable laws, and third-party rights. LinkedIn may limit or restrict accounts that use prohibited automation. The maintainers do not authorize spam, unauthorized data collection, privacy violations, or circumvention of access controls.

• Use only accounts and activity you are authorized to operate.

• The server pauses on authentication expiry, checkpoints, restriction pages, permission failures, and configuration errors.

• It does not implement CAPTCHA bypass, proxy rotation, fingerprint spoofing, credential harvesting, stealth plugins, or private LinkedIn endpoints.

• Never commit or share the persistent browser profile or local assets.

See SECURITY.md and the security design.

More documentation

• Configuration

• Capability matrix

• Architecture

• Testing

• Collection verification

• Publishing

• Changelog

Support the project

If LinkedIn MCP Server is useful to you:

• Star the repository

• Sponsor continued development

Let's connect

Have feedback or an idea for LinkedIn MCP Server?

• Connect with me on LinkedIn

• Follow me on GitHub

• Report a bug or request a feature

License

Licensed under the Apache License 2.0.