Explore without LinkedIn
Open the hosted dashboard, use the built-in records, and inspect every workflow before installing anything.
Open example dashboardFree local setup · human reviewed
Turn Claude or Codex into a LinkedIn research team.
Install four skills, connect a local read-only LinkedIn route, and turn sourced people and company research into trends, fit scores, outreach drafts, and a content map.
4Shareable skills
6Read-only tools
2Separate modes
0Automatic sends
Start here
What you are installing
A local research workflow that keeps evidence visible from the first profile through the final draft.
Research from your computer
Use your own authenticated session through a localhost bridge. Credentials and browser data stay outside the page.
See local setupLive access is conditional.The LinkedIn connector is independent and unofficial. LinkedIn can change its interface or restrict automated access. Login, CAPTCHA, and security checkpoints always remain human controlled.
Fastest setup
Paste one prompt into your coding agent
Use this in Codex, Claude Code, OpenClaw, Cursor, or another local agent that can run terminal commands.
LinkedIn Signal Analyser setup prompt
Set up the LinkedIn Signal Analyser from this repository on my computer: https://github.com/Vibha-Ramprakash/Linkedin-Manager Read README.md, SETUP-GUIDE.md, and the four SKILL.md files before changing anything. Keep the project in a new dedicated folder. Never copy credentials, cookies, browser profiles, or real lead data into the repository. Install Agent Reach from its current official GitHub guide and set up only the LinkedIn channel. Show me what is missing and ask before any system-level install or configuration write. Pause for me to complete the one-time LinkedIn login, confirmation, CAPTCHA, or security checkpoint. Never ask for my LinkedIn password or bypass a checkpoint. After login, verify the connection with a real, non-empty, read-only get_my_profile call and one people search. Configuration detection alone is not proof that LinkedIn works. Run the repository tests, start node live-server.mjs, and open http://127.0.0.1:4317. Confirm that the dashboard shows LinkedIn connected · live read only after the successful profile read. Keep every LinkedIn action serial, small-batch, and read-only. Allow only search_people, get_person_profile, get_company_profile, get_company_posts, search_posts, and close_session. Never connect, follow, message, read inboxes, publish, schedule, modify a profile, or replace an empty live result with example data. When finished, give me the Agent Reach version, LinkedIn verification result, local dashboard URL, remaining manual steps, and the account-risk warning.
Local installation
Install in five steps
Let the agent do the mechanical setup. Keep system changes and authentication under your control.
Clone the project and inspect the skills
The repository contains People Finder, Trend Scanner, Fit Checker, and Evidence Writer skill files, plus the dashboard and local bridge.
git clone https://github.com/Vibha-Ramprakash/Linkedin-Manager.gitInstall the LinkedIn research route
Follow Agent Reach's current official guide. The tested setup installs only the LinkedIn channel.
agent-reach install --env=auto --system --channels=linkedinComplete the human login
A visible browser opens. You complete sign-in and any security checkpoint yourself.
uvx mcp-server-linkedin@latest --loginVerify two harmless reads
The connection counts as working only when your profile and one people search return real, non-empty records.
mcporter call linkedin.get_my_profile sections="experience"Start the local dashboard
Serve the workspace from localhost. Do not use a file URL for live research.
node live-server.mjsYour progress
Confirm each setup milestone
This checklist is stored only in this browser. It never checks or uploads credentials.
Complete the five checks before starting a live run.
Inside the workspace
One evidence trail, four working skills
Every downstream output keeps the source, retrieval time, supported fields, and unknowns visible.
People Finder
Searches a small batch by role and market, then inspects profiles and company context.
Trend Scanner
Maps repeated questions, complaints, launches, recent activity, and wider shifts from qualified evidence.
Fit Checker
Separates ICP fit from timing and flags promotions, new roles, company moves, and launches.
Evidence Writer
Creates editable outreach drafts and a format-aware LinkedIn content schedule with source references.
Small, serial batches with provenance.
Role, company, region, and problem evidence.
Deduplicated posts and major changes.
Outreach and content, never auto-sent.
Credibility boundaries
What the system will and will not do
The setup is designed for inspectable research, not hands-off LinkedIn activity.
Research and drafting
People search, public profiles, company context, public posts, trend analysis, fit scoring, timing events, drafts, and content planning.
Account actions
No connecting, following, messaging, inbox access, publishing, scheduling, profile changes, CAPTCHA bypass, or silent example-data fallback.
Use small, serial batches.This improves inspectability and personalization, but it does not guarantee account safety or platform permission. Stop if LinkedIn or your organisation blocks the activity.
When something fails
Troubleshooting without faking success
A failed or empty live run stays failed or empty. The dashboard never fills it with example profiles.
LinkedIn connection needs attention
Keep the visible browser open, complete any login or checkpoint yourself, then repeat the same harmless profile read. A configured connector is not proof of a live session.
The search returned zero people
Keep the live result empty. Try one recognizable role, one city, and plain market language later. Do not rapidly loop searches or substitute example records.
I requested ten people but received three
Ten is a maximum request, not a guarantee. The connector exposes no pagination or result-limit parameter. Keep the returned records, then use Add to run later if you need another small batch.
Trend Scanner has fewer contributors than People Finder
Only leads with supported role and company evidence can influence themes. Profiles without dated post evidence remain available for fit, drafting, and review but stay outside trend totals.
The visible browser closed or the run was interrupted
Completed evidence remains available. Restart the local bridge, verify the live read again, and do not assume unseen steps finished.
Ready to explore
See the complete research workspace.
The hosted dashboard uses example data so you can inspect People Finder, Trend Scanner, Fit Checker, Evidence Writer, and Review Queue before connecting anything.