One MCP server,
thirty go-to-market tools
Your agent connects once. Retriever orchestrates the thirty behind it: company databases, every social platform and the live web, with the evidence that qualified each result.
Connect to any MCP-compatible tool
What your agent can do once it is connected
Find companies
On filters, or on any signal you can describe.
Find contacts
Decision makers, with verified email and phone.
Find content
Viral posts, ads, and the creators behind them.
Track signals
Hiring, funding, tech stack, ads, reviews.
Sequence them
Draft, enroll, watch the replies land.
Connect your CRM
HubSpot, Salesforce and more, coming soon.
Databases, social, and the live web
Retriever picks the right one per question: a database field when it exists, the live web when it does not.





One key, 30+ sources
Company databases, registries, ad libraries, job boards.
Live on social
Posts, threads, ads and creators, read at search time.
Live on the web
Surface, index, crawl, scrape and take action on any page.
Setup
How to install in seconds
Select your preferred AI tool below. Copy the configuration or use our hosted HTTPS endpoint to connect in seconds.
Connect to Cursor
Use Retriever tools in Cursor Composer and Chat to enrich companies and query live B2B signals.
Frequently asked questions
Which AI tools can connect to the Retriever MCP server?
Claude, Cursor, Windsurf and ChatGPT, plus anything else that speaks MCP. Copy the configuration for your tool, or point it at the hosted HTTPS endpoint.
Do I need a separate key for each source?
No. One key opens every capability, and the same key works for the MCP server and the API. You get it in the app, under Settings then API keys.
What can an agent do through the MCP server?
Find companies on criteria, find and verify contacts, read what a market publishes on social, enroll leads in a sequence and read the replies.
Does it search a database or the live web?
Both. Databases answer what they are good at, and the agent reads the live web for everything that exists in no database field, quoting the page it read it on.