We recently rebuilt the backend as a metered API with a corresponding MCP server for agents.
The original Dogpile thesis was that no single index sees the whole web. There's an old Penn State / Pitt study that found ~85% of top results were unique to one engine. Full disclosure: Dogpile sponsored it and of course, that was a totally different web. But the underlying question feels relevant again. Cursory analysis we’ve done internally shows that Brave, Google, and Bing all surface different results for the same query.
For humans, missed results are annoying.
For agents doing RAG or grounding generated content, missed or low-quality results can become a failure mode with no human in the loop to catch it.
Current state:
- Search API + MCP server (npx @system1/dogpile-mcp)
- Structured JSON output
- 3 types of responses. Basic is just what you’d get on a serp. Enhanced scrapes the result pages and includes the text as markdown. Dee performs deep research running multiple queries providing a summary.
- $0.50 monthly credit, no credit card required
- Scaling starts at $2/1k calls
Known limitations:
- It's single-source today so it is not true metasearch yet
- The roadmap is multi-source aggregation, but we wanted to get this into people’s hands as we continue to build out the feature set
- The API contract is designed so adding sources later will not require client changes
- Basic calls are landing around 800ms, but we don't have serious production concurrency numbers or p99s under heavy loads yet
Here’s a question for this group:
What features or data shapes are you missing from existing agentic search tools when building agent pipelines?
chambertime•1h ago