> ## Documentation Index > Fetch the complete documentation index at: https://docs.hivesight.so/llms.txt > Use this file to discover all available pages before exploring further. # MCP server > Query your HiveSight network in natural language from Claude, Cursor, ChatGPT, or any MCP-compatible client. HiveSight exposes a Model Context Protocol (MCP) server so you can ask your network questions in plain English from any MCP client — Claude Desktop, Cursor, the MCP Inspector, ChatGPT's MCP support, or anything else that speaks the protocol. The agent on the other end of the MCP server has access to the same tools the in-app HiveSight agent uses: warm-path lookup, account research, intro suggestions, contact lookup, ICP fit checks, and more. ## How it works Each ecosystem has its own MCP URL: ``` https://dashboard.hivesight.so/api/mcp/ ``` You can copy yours from **Settings → Integrations → MCP Server**. Authentication uses OAuth 2.1 — your MCP client follows a standard discovery handshake and signs you in with your HiveSight account on first connect. Once authorized, the client holds a refresh token and won't ask again unless you revoke access. ## Connect from Claude Desktop In Claude Desktop, click **Edit MCP server config**. Paste this entry (replace `` with yours): ```json theme={null} { "mcpServers": { "hivesight": { "url": "https://dashboard.hivesight.so/api/mcp/" } } } ``` Claude reloads the MCP server list. The first time you send a message that uses HiveSight, Claude will open a browser tab for you to sign in with your HiveSight account. ## Connect from Cursor In Cursor, open **Settings → MCP** and click **Add new MCP server**. Use the same URL from the integrations page. Cursor handles the OAuth flow automatically. ## Connect from MCP Inspector The [MCP Inspector](https://github.com/modelcontextprotocol/inspector) is useful for testing tool responses. Point it at your ecosystem URL and authenticate when prompted. ## What you can ask The agent picks the right tool from your question — you don't need to know tool names, just describe what you want. Here are the patterns that work best. Mix and match: each pattern composes with the others. ### Topic discovery — who in my network is talking about X? Surface people in your network posting or commenting on a specific topic or theme. * *"Who's been posting about AI agents in the last two weeks?"* * *"Anyone in my network commenting on Salesforce vs HubSpot?"* * *"Show me people writing about Series A fundraising recently."* ### ICP-fit thought-leadership monitoring — posters who don't get a lot of engagement ICP-fit people who post regularly but get modest engagement are high-leverage outreach targets — active, in your buyer profile, and not yet over-saturated by inbound. * *"Find ICP-fit people posting under-the-radar content I could amplify."* * *"VP Sales people in my network posting weekly but getting under 20 reactions — they'd notice an inbound."* * *"ICP creators with high post volume and low engagement, sorted by post count."* ### Trigger-based workflows — whenever X happens, do Y MCP doesn't run cron, but you can replay these on a cadence (a Monday morning DM works well) and chain detection with a follow-up action like drafting a message or adding a connector. * *"Whenever someone in my ICP changes jobs, write me a congrats message."* * *"When a connector engages on a post by an ICP-fit person, surface the post and draft an intro ask."* * *"If anyone in my network announces a Series B, add their VP Sales as a target contact."* ### Account momentum — what accounts are heating up? Spot accounts where multiple employees are interacting with your voice — a real buying signal. * *"Which target accounts have multiple employees engaging with my network this month?"* * *"Companies where 2+ people are interacting with my connectors lately."* * *"Rank my target accounts by recent engagement volume."* ### Bridge-finder before cold outreach — who knows them? Always check for a warm path before going cold. Even weak signals (work overlap, recommendations) can be worth a name-drop. * *"I want to reach the VP Sales at Acme — who's my warmest bridge?"* * *"Best path into Stripe's new Head of GTM?"* * *"Strongest evidence anyone on my team knows Jane Doe."* ### Dormant-relationship reactivation — who have I gone quiet on? Catch lapsed warm connections before they go cold. * *"Connectors I haven't met with in 60+ days who still have strong relationships."* * *"Lapsed customers who used to be active in our network."* * *"Top 10 warm relationships I should re-engage this month."* ### Engagement-pattern intel — how do these two people actually know each other? Before asking a connector for an intro, understand the basis of the relationship. * *"What's the history between Stephen and Jane? Should I name-drop or ask for a direct intro?"* * *"Strongest evidence Stephen knows Jane warmly."* * *"Have they actually interacted recently, or is this just a LinkedIn connection?"* ### Post-meeting extraction — I just got off a call, what next? Calendar-aware. Pulls accounts you've actually met with and ranks intro paths by ICP fit. * *"I met with someone from Coupa, what intros can I extract?"* * *"Best ICP-fit paths from my Monday meetings."* * *"Rank my recent accounts by intro opportunity."* ### Add a connector on the fly Maintain your network without leaving your AI client. Works as the follow-up step in trigger workflows. * *"Add Casey Nolan as a customer connector."* * *"After this meeting, mark the attendee as a partner connector."* ### ICP coverage audit — where are the gaps? Find target accounts where you have no warm path. These are where you need to either add new connectors or accept the account is cold. * *"Which target accounts have zero ICP-fit intro paths?"* * *"Show target companies where I don't have a single warm path into a decision-maker."* **Tips for getting good answers.** Be specific about time windows — "last two weeks" / "this month" gives the agent a clear filter. Name connectors by name when you can ("through Stephen" beats a generic ask). If the first answer is too broad, narrow with a follow-up: *"only VPs"*, *"exclude my customers"*, *"only at companies under 200 people."* Each MCP URL is scoped to one ecosystem. If you have multiple ecosystems, add one MCP server per ecosystem in your client config with distinct names — `hivesight-enterprise`, `hivesight-mid-market`, etc. ## Security notes * The MCP URL itself isn't a secret — the OAuth handshake is what authorizes access. Anyone who hits the URL without a valid token gets a 401. * Tokens are scoped to your HiveSight user. If you're a member of a team ecosystem, your MCP tokens only see the ecosystems you have access to. * Revoke a connected client at any time by removing the HiveSight MCP entry from the client's config (Claude Desktop, Cursor, etc.) — the refresh token stops working as soon as it's removed.