What bouncewatch/mcp MCP server does
bouncewatch/mcp MCP server gives an AI client access to Bounce Watch’s company-event data. The service covers more than forty signal types, including funding, senior hires, office openings, customer wins, partnerships, and expansion. Each signal includes the date the event occurred, rather than only the date it was discovered, and carries a weight from 1 to 10.
The service supports natural-language questions about recent activity, company characteristics, and account changes. Results include company names, dates, and sources. A quiet result should be interpreted with the reported coverage recency: an empty response does not necessarily mean that no event occurred.
How it works
The hosted MCP endpoint is available at https://api.bouncewatch.com/api/v1/mcp and uses streamable HTTP. The repository contains connection details, client examples, and a launcher; it does not contain a local copy of the data service. The Dockerfile likewise packages the launcher rather than the hosted server.
Authentication uses OAuth or an API key supplied through X-API-Key or Authorization: Bearer. The handshake, available tools, and prompts can be viewed without credentials, but running a tool requires authentication. New accounts receive 2,500 credits, and a typical question costs about 50 credits. Taxonomy lookup and refresh-status checks do not consume credits.
Five prompts are included: pre_round_radar, why_now, funded_and_hiring, account_watch_brief, and risk_scan. They cover growth signals, outreach context, recent funding and hiring, account changes, and company risk events.
Setup and configuration
For clients that support remote HTTP MCP connections, configure the endpoint directly. Claude Code can add it with its HTTP transport command, while Cursor and VS Code examples are provided in the repository. Claude Desktop should use its custom connector interface for the remote URL rather than its local-server configuration file.
Clients limited to stdio can use the published @bouncewatch/mcp launcher through npx. Set BOUNCEWATCH_API_KEY in the launcher environment, and the process forwards requests to the same hosted endpoint. n8n and Make can connect by entering the remote URL and adding the API key as a request header.
Tools and capabilities
The ten tools are:
search_signals: find recent signals across the indexed companies.
search_companies: filter companies by country, headcount, and funding stage.
find_company: resolve a company name to its domain.
get_company: retrieve a company’s firmographic profile.
get_company_signals: read a company’s dated signal timeline.
refresh_company: queue a new scan for a company.
get_refresh_status: inspect a queued scan.
watch_company: create a standing watch for a company.
check_watches: review events for companies already watched by the key.
get_signal_taxonomy: list signal types grouped by category.
refresh_company and watch_company modify state; the other eight tools are read-only.
Limitations and notes
bouncewatch/mcp MCP server is a connection to a commercial hosted service, not a self-contained data server. Tool use requires an account and authentication, and usage is credit-based except for the two free lookup operations described above. The repository’s launcher is intended for clients that cannot speak streamable HTTP; clients with remote MCP support can connect directly instead.
The service reports when company coverage was last checked, which matters when interpreting missing results. The repository is licensed under MIT, while the hosted Bounce Watch service remains a commercial product.