remote-mcp-linkedin

Disclaimer: This is an independent, community project. It is not affiliated with, authorized by, endorsed by, or sponsored by LinkedIn Corporation or Microsoft. "LinkedIn" is a registered trademark of LinkedIn Corporation and is used here only descriptively to identify the third-party service this software interoperates with.

remote-mcp-linkedin is a Remote LinkedIn MCP server with a local browser bridge. The server exposes read-only MCP tools and builds structured dossiers; the bridge runs on the user's machine, uses the user's local browser session, and returns only visible extracted profile data.

Cookies, browser profiles, storage state, auth headers, session state, browser fingerprints, and CDP access never leave the local bridge machine.

v0.1 Capabilities

v0.1 is single-user and single-bridge. The architecture leaves room for multi-user routing later, but it is not implemented in this release.

Related MCP server: LinkedIn MCP

Architecture

flowchart LR
  Client["MCP client"] -->|MCP stdio by default| Server["remote-mcp-linkedin server"]
  Server -->|read-only command| Manager["Bridge session manager"]
  Bridge["local browser bridge"] -->|outbound authenticated WebSocket| Manager
  Bridge -->|local only| Browser["Local Chromium / Chrome / Patchright profile"]
  Manager -->|raw visible sections| Server
  Server --> Builder["Dossier builder"]
  Server --> Store["Local JSON result store"]
  Server -->|JSON result| Client

The server is the MCP and orchestration layer. It does not launch a browser. The bridge is intentionally thin: it receives allowlisted read-only commands, extracts visible browser data locally, and returns raw sections.

Quickstart

Prerequisites: Python 3.12+, Git, and a local shell.

1. Install From Source

git clone https://github.com/alexanderskorokhodov/remote-mcp-linkedin.git
cd remote-mcp-linkedin

python3.12 -m venv .venv
source .venv/bin/activate
python -m pip install -e ".[dev]"

If your machine has a newer Python available, python3.13 or python3.14 is also fine.

2. Start The Server

Set a long random token. The same value must be used by the server and bridge.

export REMOTE_MCP_LINKEDIN_BRIDGE_TOKEN="replace-with-a-long-random-token"
remote-mcp-linkedin-server

The server uses MCP stdio by default. It also starts a local WebSocket listener for the bridge on 127.0.0.1:8765.

3. Start The Local Bridge

In another terminal:

cd remote-mcp-linkedin
source .venv/bin/activate

export REMOTE_MCP_LINKEDIN_BRIDGE_TOKEN="replace-with-a-long-random-token"
remote-mcp-linkedin-bridge

By default, the bridge uses the stub extractor. This is expected in v0.1 and is useful for validating the full MCP/server/bridge flow without opening a real LinkedIn session.

4. Optional HTTP MCP Mode

HTTP is disabled by default because v0.1 does not ship MCP HTTP authentication. Only enable it on loopback or behind your own trusted access control.

remote-mcp-linkedin-server \
  --transport streamable-http \
  --enable-http \
  --mcp-host 127.0.0.1 \
  --mcp-port 8000 \
  --mcp-path /mcp

MCP Client Configuration

For a local MCP client that launches the installed console script:

{
  "mcpServers": {
    "remote-mcp-linkedin": {
      "command": "remote-mcp-linkedin-server",
      "env": {
        "REMOTE_MCP_LINKEDIN_BRIDGE_TOKEN": "replace-with-a-long-random-token"
      }
    }
  }
}

Run remote-mcp-linkedin-bridge separately on the machine that owns the browser session.

MCP Tools

Available sections:

top_card, about, experience, education, skills, certifications,
projects, languages, contact_info, posts

Experimental Browser Extraction

The default extractor is stub, which returns deterministic mock data. This is not hidden: responses include a STUB_BROWSER_EXTRACTION warning.

To try the experimental Patchright extractor:

python -m pip install -e ".[browser]"

export REMOTE_MCP_LINKEDIN_EXTRACTOR=patchright
export REMOTE_MCP_LINKEDIN_HEADLESS=false
remote-mcp-linkedin-bridge

Patchright uses a local persistent browser profile under:

~/.remote-mcp-linkedin/bridge-profile

Override it with REMOTE_MCP_LINKEDIN_BROWSER_USER_DATA_DIR if needed. That path and its contents are local-only and must not be sent to the server.

Configuration

Security Model

• The bridge authenticates to the server with a shared token.

• The bridge makes an outbound WebSocket connection; the server does not reach into the user's browser.

• The server rejects unauthenticated bridge connections.

• The MCP server uses stdio by default and refuses HTTP unless --enable-http is explicitly provided.

• The bridge accepts only allowlisted read-only commands in v0.1.

• Cookies, browser profiles, storage state, auth headers, full session state, fingerprint material, and CDP access are never sent to the server.

• Logs must not include bridge tokens, cookies, auth headers, browser profile paths, or browser session state.

Read more in docs/SECURITY.md.

Limitations

• LinkedIn page structure changes frequently; DOM/text extraction is fragile.

• Use may be subject to LinkedIn terms and account restrictions.

• Extracted data is limited to what the local logged-in user can see.

• Captchas, login challenges, rate limits, and account checkpoints can block extraction.

• v0.1 is single-user/single-bridge.

• The real browser extractor is intentionally minimal and incomplete.

Development

python -m pip install -e ".[dev]"
pytest
ruff check .

Terminal 1:

export REMOTE_MCP_LINKEDIN_BRIDGE_TOKEN="dev-token"
remote-mcp-linkedin-server --transport streamable-http --enable-http

Terminal 2:

export REMOTE_MCP_LINKEDIN_BRIDGE_TOKEN="dev-token"
remote-mcp-linkedin-bridge --extractor stub

Connect an MCP inspector or FastMCP client to:

http://127.0.0.1:8000/mcp

Documentation

• Architecture

• Security

• Bridge protocol

• Dossier schema

• Development

Attribution

This project was designed with reference to stickerdaniel/linkedin-mcp-server, which is licensed under Apache-2.0. remote-mcp-linkedin uses a separate remote server/local bridge architecture and intentionally omits write-action tools in v0.1.