LinkedIn Toolkit

Operator, Browser Use, computer-use models and Playwright bots get challenged or banned on LinkedIn: headless fingerprints, datacenter IPs, machine-speed clicks. LinkedIn Toolkit gives any agent a safe, structured API to your own logged-in Chrome session — through the same internal endpoints the LinkedIn page itself calls, at human pace, under hard caps, with a human approval queue. It is also a free replacement for Waalaxy and PhantomBuster if you never touch an agent at all. No headless browser, no proxies, no cloud session, no telemetry, no subscription.

Quick start

One command, then three clicks in Chrome.

npx linkedin-toolkit-mcp setup --client claude-code

That downloads the extension for your version, checks it, unpacks it to ~/.linkedin-toolkit/extension, prints the pairing token, waits for the extension to connect, and writes the MCP config block into your client's own config file — keeping every other server in it and backing the file up first. --client takes claude-desktop, claude-code, cursor, windsurf, vscode, n8n or print; add --dry-run to see what it would do without touching anything.

Then the three steps Chrome does not let any installer do for you:

1. Open chrome://extensions — type it in the address bar; it does not come up in search.
2. Turn on Developer mode, top right.
3. Click Load unpacked and choose the folder setup printed.

Paste the pairing token into the popup → Settings → Local bridge, and check it took:

lit status

Mass unfollow — Preview first, then a small number, then the rest — is in docs/install.md. Read the "Connections are followed too" part before you decide the feature is broken: LinkedIn's Following list does not include your connections, who are followed automatically when you connect, so emptying that list to zero leaves a feed still full of posts. Ticking "Also unfollow my connections" scans your followers list instead, which is the only place that state is visible.

Every client's config file, path and shape: docs/clients.md · one page per ecosystem.

Not using MCP? lit serve --http gives you POST /actions/{action} and a generated GET /openapi.json. Examples in seven languages.

[!IMPORTANT] Nothing sends without you. Copilot mode is the default: every write an agent makes queues for your approval in the popup. Hard caps live in the extension — 100 invites, 150 messages, 500 profile visits, 1,000 search results a day — and no agent, CLI flag or config file can raise them. Read the safety page before you turn Autopilot on.

What it does

Works with your agent

Structured errors carry code, message, retryAfter and howToFix, so an agent recovers or explains itself instead of retrying into a wall. There is an llms.txt and an agent quickstart written for an agent to read and self-install.

Browser agents vs LinkedIn Toolkit

The long version, with the actual detection mechanisms: Why browser agents fail on LinkedIn.

Why extensions broke, and why this one is built to be repaired

LinkedIn's web client now serves nearly all of its data through GET /voyager/api/graphql?queryId=<name>.<32-hex hash>&variables=(...), and those hashes change with each web client release (current: 1.13.46474). The old REST Voyager paths that a generation of 2024–2025 extensions hard-coded return 400, 410 or 500 today. That is the mechanism — not a ban wave. This extension calls the same GraphQL queries the page calls, from inside your own tab, and keeps every query ID in one refreshable table with its capture date and client version: docs/voyager-endpoints.md. lit endpoints check reports which are ok, failed or unverified, and lit endpoints doctor names the query id whose hash went stale and the key in the table that holds it, so drift is a maintenance task rather than an architecture change.

Honestly: those IDs will drift, and re-capturing them is the contribution this project most needs. It is a table edit, not a rewrite — open DevTools on LinkedIn, filter the Network tab for voyager/api, and copy the queryId from a request the page makes; the same hashes are also literal strings inside LinkedIn's JS bundles if you would rather grep for them.

Versus the paid tools

_{Competitor prices are public list prices checked September 2026 and are approximate — they change, vary by currency and billing term, and each vendor's tiers differ. Feature claims are taken from each vendor's public product pages, also checked September 2026, and tiers move. Check their sites before deciding anything. Corrections welcome via PR — if we have a feature wrong, open one and it gets fixed.
Source for the Waalaxy column: its current Chrome extension listing, "Alien Copilot" by Waapi (Montpellier) — v1.1.3, updated August 2026, roughly 2,000 users, 3.0★ from 3 ratings — which describes itself as "your Waalaxy companion, helps you import prospects". On that listing the extension imports prospects into Waalaxy, and the automation runs on Waalaxy's servers using your session. Listing details read September 2026.}

Research Pack

Drop in a CSV with any of name, linkedin_url, email, domain, company. Get back a dossier per row, an enriched CSV, and a list — all local.

lit research leads.csv --out ./packs

1. Resolve — match each row to a profile or company. Ambiguous rows come back with candidates and a confidence score for you to pick from, rather than a silent guess.
2. Gather — full profile capture, company page, recent posts and engagement, mutual connections, connection status.
3. Signals — job change in the last 90 days, recent posting activity, hiring signals, headcount band, mutuals, engaged-with-me.
4. Enrich (optional, your key, off by default) — verified email and phone.
5. Web — the linkedin-research-pack skill has your agent use its own web search for news, talks, GitHub and podcasts, and write them into the pack with sources. The extension never crawls the open web.
6. Write — pack.md and pack.json per row, an output.csv with every original column plus resolved URL, title, company, location, signals and match confidence, and a new list.

Caps apply throughout: resolution spends search quota, capture spends visit quota. A 500-row CSV is a multi-day job by design, and you get the ETA up front.

Safety

The honest position: LinkedIn's User Agreement prohibits automated access. This tool automates LinkedIn. Nothing below makes that risk zero.

What it does do:

• Hard caps in the extension, below every client: 100 invites, 150 messages, 500 profile visits, 1,000 search results per day. config.set clamps whatever you pass.
• Human pacing — jittered 8–15 second delays, hourly caps, a business-hours window, weekdays only if you want. Machine-speed activity is the loudest signal an account can emit.
• 14-day warm-up for new or dormant accounts.
• Copilot mode — every agent write queues for your approval. Autopilot is a toggle only a human can flip, in the popup. Approving still is not sending: the engine paces it anyway.
• 429 → backoff. A security check → stop. A CAPTCHA or "confirm it's you" check — whether it arrives as a 451 or, as it usually does, as a redirect to LinkedIn's checkpoint page — pauses every request immediately and stays paused until you have completed it yourself and said so in the popup. There is no retry loop anywhere in the codebase.
• Never bypasses a security measure. No CAPTCHA solving, no challenge circumvention, no proxies, no fingerprint spoofing, no cookie import, no account you are not signed into. What triggers a check, what the toolkit does, what you should do and what we will not build: docs/captcha-and-security-checks.md.

What it does not do is make you invisible. Running inside your own session removes the fingerprint and IP signals that get browser agents caught — it does nothing about how much you do or how regularly you do it, and LinkedIn counts both. That is exactly why the caps and the pacing are not configurable past a ceiling: they are the only defence left once the easy tells are gone. An account sending 90 invites a day at perfectly spaced intervals is still an account sending 90 invites a day.

Recommended settings, signs to stop, and your data-protection obligations: docs/safety.md.

Architecture

flowchart LR    A["Your agent<br/>Claude · Cursor · LangChain<br/>CrewAI · n8n · curl"]    M["linkedin-toolkit-mcp<br/><i>your machine</i><br/>MCP · HTTP · SQLite · CLI"]    E["Extension engine<br/><i>your Chrome</i><br/>quotas · delays · queue<br/>campaigns · lists"]    Q["Approval queue<br/><i>you</i>"]    L["LinkedIn<br/><i>your session, your cookies,<br/>your IP, your device</i>"]
    A -->|"MCP stdio / HTTP"| M    M <-->|"ws://127.0.0.1:47829"| E    E --> Q    Q -->|"you approve"| E    E -->|"Voyager API, human pace"| L

One engine, several clients: the popup, the CLI, an MCP tool call and a campaign step all go through the same handle(action, params, origin) switch. The caps and the queue sit below it, so there is no path around them — there is only one path. Full architecture · action contract · tool reference.

CLI

lit ships in the same npm package as the server.

lit setup --client cursor                                # install, pair, configure. Start herelit statuslit search "CTO fintech London" --source salesnav --count 100 --csv out.csvlit profile https://www.linkedin.com/in/... --full --jsonlit engagers <post-url> --list "Post engagers 8 Sep"lit invite <profile-url> --note "..."                    # queues in Copilot modelit campaign create --from sequences/warm-connect.json --list "Data leads"lit inbox --since 24h --sentimentlit research leads.csv --out ./packslit sql "select company, count(*) from profiles group by 1 order by 2 desc limit 20"lit export --table profiles --csvlit serve --http                                         # HTTP MCP + /actions + /openapi.json

Full reference · shell examples.

Your data is a SQLite file

Everything you capture mirrors into ~/.linkedin-toolkit/toolkit.db. Agents get read-only SQL over it — no network, no quota, no rate limit — and you can open the same file in any SQLite tool.

-- Who accepted an invite but never repliedSELECT p.full_name, p.company, p.headline, a.created_atFROM actions aJOIN profiles p ON p.public_id = a.public_idWHERE a.action = 'outreach.invite' AND a.accepted = 1  AND p.public_id NOT IN (SELECT from_public_id FROM messages)ORDER BY a.created_at DESC;

lit sql "select company, count(*) n from profiles group by 1 order by n desc limit 20"

An agent reaches the same thing through linkedin_query_sql with { "sql": "SELECT …" }. SELECT only — anything else is rejected.

Webhooks

The server POSTs { event, payload } to a URL you set — invite_accepted, reply_received, positive_reply, campaign_step_done, campaign_completed, quota_hit, challenge_detected, queue_item_added, queue_item_sent, campaign_note_truncated, research_progress, research_completed.

lit config set webhookUrl https://your-n8n/webhook/linkedin-events

An importable n8n workflow does the obvious thing with them: accepted invite → profile.get → an LLM drafts a first message → it queues → Slack asks a human → approval link → queue.approve.

Skills

Six task recipes in the Agent Skills format. They carry the guardrails — facts only, quota awareness, the approval queue as the expected destination — not just the tool sequence.

cp -r skills/* ~/.claude/skills/        # or ~/.openclaw/skills/, or ./.claude/skills/

linkedin-sourcer · linkedin-outreach-writer · linkedin-campaign-runner · linkedin-profile-to-dossier · linkedin-reply-triage · linkedin-research-pack

Roadmap

What is coming, and the things that will never be built because they need a server or break the local-first guarantee: docs/roadmap.md.

Contributing

Adding an extractor is the best first contribution and touches four files: how to build one. Sequences, skills and agent integrations are merged fastest because they are additive and self-contained.

See CONTRIBUTING.md, pick up a good first issue, or open a Discussion.

Disclaimer

THIS SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND. The authors and contributors accept no responsibility or liability for any consequences arising from its use, including but not limited to:

• LinkedIn account restrictions, suspensions, or permanent bans
• Loss of connections, data, or account access
• Violation of LinkedIn's Terms of Service or User Agreement
• Any direct, indirect, incidental, or consequential damages

By using this software you acknowledge that:

1. LinkedIn's User Agreement prohibits automated tools and scraping, and using this may breach it
2. Doing so may result in action against your LinkedIn account, up to permanent loss
3. You use it only on your own account, in a session you logged into yourself
4. You are solely responsible for every action taken with it, and for your obligations under GDPR, the UK GDPR, CCPA or any equivalent law covering the personal data you collect
5. You use it entirely at your own risk

This tool never bypasses a security measure: no CAPTCHA solving, no challenge circumvention, no detection evasion, no proxies, no cookie theft, no session sharing, no accounts you are not signed into. When LinkedIn puts up a wall, it stops and hands the problem to you.

Provided for educational and research purposes. We do not encourage or endorse violation of any platform's terms of service.

License

MIT.

Contributors

This project uses all-contributors. Contributions of any kind are recognised here — code, docs, sequences, skills, bug reports, and design.

To add someone, comment on any issue or PR:

@all-contributors please add @username for code, doc

Credits

Built by Dominic Gonsalves — LinkedIn · GitHub

If it is useful, a star helps other people find it.

[Star History Chart]