# ResearchHub Foundation Science Radar

> ResearchHub Foundation provides reviewed research workflows and source-linked Trending Science signals through a public MCP server for compatible research agents.

## Start here

- [Latest Science Radar](https://radar.researchhub.foundation/radar): Browse the latest public science signals, full public history, attention context, and retained source links.
- [Radar use cases](https://radar.researchhub.foundation/use-cases): Source-linked workflows for research intelligence, science communications, portfolio context, policy/program horizon scanning, and AI research agents.
- [MCP Server for Research Workflows and Science Radar](https://radar.researchhub.foundation/mcp): Connection guidance for reviewed manuscript/proposal workflows and source-linked public Radar data.
- [Agent entrypoint](https://radar.researchhub.foundation/agents.md): Concise, credential-free instructions for agents visiting this website.
- [A2A agent card](https://radar.researchhub.foundation/.well-known/agent-card.json): Public A2A v1 discovery card for a guidance-only Science Radar concierge.

## MCP client guide

- [Impact MCP client-side agent guide](https://radar.researchhub.foundation/impact-mcp-agents.md): Full setup guidance for Codex, Claude, and compatible MCP clients; available tools; safe workflows; and troubleshooting.
- [Portable Impact MCP Search Benchmark](https://radar.researchhub.foundation/benchmark/benchmark-instructions.md): Harness-neutral instructions for comparing Direct Search with Impact MCP Search across the frozen prompt set.
- [Benchmark manifest](https://radar.researchhub.foundation/benchmark/benchmark-manifest.v1.json): Machine-readable prompts, lane boundaries, metadata contract, repeats, and judge protocol.
- [Remote harness execution prompt](https://radar.researchhub.foundation/benchmark/harness-execution-prompt.md): Provider-neutral instructions for running the benchmark and returning raw results for operator-side JEV judging.
- [Benchmark submission contract](https://radar.researchhub.foundation/benchmark/benchmark-submissions.md): Run creation, bounded batch upload, finalization, and operator retrieval semantics.
- Canonical public MCP endpoint: `https://mcp.researchhub.foundation/mcp`. No key is needed while public mode is enabled.
- Active workflow catalog: `impact://skills/catalog`; complete skill template: `impact://skills/{skill_id}/SKILL.md`; tool fallback: `get_impact_record` with the listed skill document ID.

## Data and citation boundary

- [AI access note](https://radar.researchhub.foundation/ai.txt): What these public resources are and are not.
- The Radar's attention labels describe public coverage and recency, not scientific quality, medical validity, clinical efficacy, research priority, or funding availability.
- For a material finding, retrieve the complete public signal and review its linked sources. Verify current facts against primary sources.
- The A2A discovery agent and `agents.md` are public guidance services. Impact MCP serves its bounded public projection without a key.