MCP server
Connect your agent.
Hyperresearch supplies evidence and research tools. Conversation stays inside the AI agent you choose and pay for. Point MCP-capable clients at:
https://mcp.hyperresearch.ai/mcpThe client discovers /.well-known/oauth-protected-resource/mcp, registers its
callback, and opens browser PKCE authorization at the API origin. The consent
page binds the selected workspace and exact scopes. It lists every scope the
client requested, all ticked, and the grant carries only what stays ticked; a
client that requests none gets mcp vault:read runs:read. Writing or metered
work needs vault:write, runs:write, fetch, or verify. Revoke a connection
from Console, Connect, or with CLI logout.
Run the CLI from source
Section titled “Run the CLI from source”The CLI is built and tested in the private source tree. It is not published to npm yet.
pnpm --filter @hyperresearch/cloud-cli buildnode apps/cli/dist/main.js loginnode apps/cli/dist/main.js status --jsonnode apps/cli/dist/main.js vault pull --output ./my-vaultnode apps/cli/dist/main.js vault push ./my-vault --waitnode apps/cli/dist/main.js mcpAnswers
Section titled “Answers”Answers is a separate MCP server at https://answers.hyperresearch.ai/mcp
with two tools: ask returns one short answer with checked quotes, and
search returns ranked sources. See Answers. The same tools are
packaged for code:
- Answers in the Vercel AI SDK (
@hyperresearch/ai-sdk). - Answers in LangChain, Python (
langchain-hyperresearch) and JavaScript (@hyperresearch/langchain).
Further reading
Section titled “Further reading”- Hyperresearch in Claude, in Cursor and in ChatGPT: client-by-client setup.
- Long-running runs over MCP: why a run starts and polls rather than blocking a tool call.