Model Context Protocol

LinkedIn intelligence for AI agents

Explore people, topics, and networks on LinkedIn from Claude, Cursor, or any MCP-compatible agent.

Start in your browser Connect your agent

Connector URL: https://fluxgraph.app/mcp

What it is

A LinkedIn data layer your agent can call.

Flux is a remote MCP server. You connect it once, and your agent gains 63tools that read LinkedIn: any public profile’s posts and reactions, semantic search across a post library, and an engagement graph built from who reacted and commented on what.

No scraping setup, no CSV exports, no browser extension. Your agent asks a question in the conversation you’re already having, calls the tools it needs, and answers with real data instead of a guess.

The tools

Three dimensions: people, topics, networks.

LinkedIn is a graph of people posting about topics and reacting to each other. The 63 tools split along exactly those three axes.

People

What a person publishes, and how it lands.

Point an agent at any public LinkedIn profile — yours, a client’s, or someone you want to learn from — and get their posts, their metrics, and the patterns underneath them.

linkedin_profile_posts: Recent posts from any public person profile, pulled live.
linkedin_profile_info: Resolve a LinkedIn URL into name, headline, company, and work history.
get_posts: Every post in your library with full metrics and text, filterable by date, type, or engagement.
get_profile_summary: KPI summary: volume, average engagement, trend, top format, goal status.
get_posting_insights: Posting patterns — cadence, best day and hour, which formats carry.
get_engagement_trends: Engagement over time, so you can see where it broke and when.
linkedin_profile_reactions: What a person has been reacting to lately — their attention, not their output.
linkedin_profile_comments: What a person has been commenting on, and what they said.
linkedin_people_search: Find people by company, title, location, or keyword.

Topics & conversations

What a subject actually sounds like on LinkedIn.

Semantic search over your own library and live keyword search across LinkedIn, plus the comment threads — so an agent can tell you what a topic is doing, not just that it exists.

search_posts: Hybrid keyword + semantic search across everything you track.
find_similar_posts: Nearest-neighbour search on embeddings — find the posts closest to a passage of text.
get_topic_insights: Which themes you post about, and how each one performs.
linkedin_post_search: Search LinkedIn posts by keyword, live.
linkedin_post_comments: Who commented on a post, and what they wrote.
linkedin_company_posts: Recent posts from any company page.
suggest_angles: Five concrete angles on a topic, grounded in what already works in your data.
score_draft: Predicted engagement for a draft, before you publish it.

Networks

Who engaged, where they work, and how you reach them.

Reactions and comments become a graph. Ask who showed up on a post, which of them sit inside a target account, and what the shortest warm path to a person is.

linkedin_post_reactions: The people who reacted to a post — persisted into your engagement graph.
get_post_engagers_by_company: Group a post’s engagers by current employer.
get_post_set_engagers: Deduped engagers across a whole set of posts, not just one.
get_engagement_industry_breakdown: Engagers bucketed by industry, with sample companies.
get_alumni_engagers: Engagers who are current employees or alumni of a target company.
get_company_engagement_overlap: People who engaged with both Company A’s and Company B’s posts.
get_outreach_path: Warm-intro paths to a target person via coworker and alumni edges, up to three hops.
score_connections: Score every connection in a campaign and rewrite the queue order.
get_outreach_queue: The prioritized queue — who to reach, in what order, with what opener.
get_target_warmth_brief: A warmth score for one person against your posts, with the posts they touched.
map_constellation: Map the whole graph around a target account in one call.

Plus streams, collections, segments, alerts, credit accounting, and a sandboxed traverse_graph escape hatch for read-only SQL. Full tool reference.

What people ask

These are real questions, in customers’ words.

Every one of these resolves to real tool calls. Paste a LinkedIn profile or post URL in place of the placeholder and it runs as written.

“What are the top performing posts of <person> so I can study their content?”

Pulls their recent posts live, ranks them by engagement, and reads back the actual text so you can see what the pattern is.

linkedin_profile_postsget_posts

“What has <person> been paying attention to lately?”

Their reactions and comments — what they read and respond to, which is usually a better signal than what they publish.

linkedin_profile_reactionslinkedin_profile_comments

“Get the data for my client's LinkedIn posts in one place”

Every tracked profile, every post, every metric — returned as structured data your agent can summarize, chart, or drop into a report.

list_profilesget_postsget_profile_summary

“Who engaged with this post and which of them work at target accounts?”

Fetches the reactors and commenters, groups them by current employer, and flags the ones inside companies you care about.

linkedin_post_reactionsget_post_engagers_by_companyget_alumni_engagers

Setup · 60 seconds

Connect your agent.

One endpoint, one bearer token. Generate a key from Settings → MCP, then pick your client.

Claude

Claude Code, Claude Desktop, Cursor, Zed — one command or one JSON block.

Set up Claude →

ChatGPT & Codex

Codex CLI and Codex in VS Code connect today with a bearer key. ChatGPT web connectors are OAuth-only — read the current state.

Set up Codex →

Anything else that speaks MCP over HTTP works too — point it at https://fluxgraph.app/mcp with an Authorization: Bearer header.

Pricing

MCP access is on Edge.

The connector is part of Edge — $15/month, or $150/year. That includes the dashboard, the scoring model, semantic search, the engagement graph, and all 63 MCP tools against your own data.

Most tools are free to call. The ones that reach out to LinkedIn live (any linkedin_* tool, suggest_angles, map_constellation) draw from a monthly credit allowance. Your agent can call get_cost_estimate before anything expensive and get_credit_balance to check what’s left.

FAQ

Questions people ask first.

What is the connector URL?

https://fluxgraph.app/mcp — a single remote MCP endpoint over Streamable HTTP. Auth is a bearer token: one API key per user, generated from Settings → MCP. Keys start with flux_ and are shown exactly once.

Which agents can connect?

Anything that speaks MCP over HTTP and lets you set a request header. Verified today: Claude Code, Claude Desktop, Cursor, Zed, and OpenAI’s Codex CLI. Browser-based connector directories (claude.ai and ChatGPT web) currently accept OAuth or no auth rather than a bearer key, so use a desktop or CLI client for now — see the ChatGPT & Codex page for the detail.

Whose data can my agent see?

Your own library, plus whatever public LinkedIn data you ask it to pull. The API key scopes every request to your account — no other user’s data is reachable with your key. Data only leaves Flux when you prompt the agent to read it.

Do I need to use an agent at all?

No. Everything the connector exposes also has a dashboard, and most people start there. The connector is for the days you’d rather ask a question than open a tab.

What does it cost?

MCP access is on the Edge plan — $15/month or $150/year. Analytics, search, scoring, and engagement-graph queries are free to call; live LinkedIn pulls draw from a monthly credit allowance. See pricing.

One ask away.

Start in the browser, or wire it into the agent you already work in.

Start in your browser Connect your agent