Rivul MCP and REST API for AI Agent Research

A research workflow often starts in one tool and needs evidence in another. A developer may need paper metadata in a service; a researcher may want an AI client to search for relevant work without leaving the conversation. Rivul Rivul MCP gives both workflows a shared, read-only scholarly search layer while keeping access and usage attached to a Rivul workspace.

In an AI agent research workflow, Rivul MCP can discover research papers and return citation metadata such as authors, publication year, DOI, citation count, and source locations when the record supplies them. It does not generate a finished bibliography or decide whether a citation supports the agent's claim.

Rivul MCP is available to confirmed accounts within published usage limits. The initial scope is deliberately narrow: search scholarly records, return useful identifiers and source locations, and report current usage. Rivul MCP does not host publishers' full text, judge whether a paper is correct, or replace reading the original source.

REST API and MCP server

The REST API and MCP server reach the same normalized search capability, but they serve different callers. Treating them as separate interfaces keeps each setup understandable and avoids asking an AI client to imitate an application integration, or the reverse.

The Model Context Protocol overview describes the common host, client, and server roles. Rivul MCP acts as a remote MCP server. Its first tool, search_papers, returns scholarly records;get_paper_details returns the full record for one work; andusage_status reports the workspace's MCP usage for the current UTC day. The model receives only the tool contract and result needed for that interaction, not Rivul's internal search infrastructure.

Two ways to authorize an MCP connection

MCP support does not mean every client presents the same setup screen. Rivul MCP therefore offers two access paths and shows the relevant one inside each client guide.

Both paths were tested against a real connection in Claude Code: the Claude Code walkthrough covers OAuth and dedicated-key access, tool discovery, a bounded scholarly search, a usage read, and revocation. The OpenAI Codex walkthrough records the same sequence for browser authorization only; key mode was not tested in Codex. Start from the one matching the client you already use.

Official MCP references by client

These are the vendors' own MCP references. They explain each client's current configuration and behavior; Rivul MCP in the app supplies the matching values. Listing a client here identifies a setup path. It does not claim a partnership, endorsement, marketplace listing, or universal OAuth certification.

Claude CodeAnthropicCodexOpenAIMCPModel Context ProtocolCursorCursorGoogle AntigravityGoogleMuse CodeMeta

Claude Code, OpenAI Codex, Cursor, Google Antigravity, and Muse Code each have a Rivul MCP setup guide, and any client that supports remote MCP servers can connect the same way. Across those paths, Rivul MCP exposes one Streamable HTTP endpoint at https://rivul.ai/mcp and three read-only tools: search_papers, get_paper_details, and usage_status.

For example, use the Claude Code MCP guide for Claude's connection commands, the Codex MCP guide for OpenAI's current client behavior, the Google Antigravity MCP guide for Google's setup, the Cursor MCP guide for Cursor's configuration, and the Muse Code MCP guide for Meta's settings shape. Those pages can change as the clients evolve, so Rivul links to them instead of copying vendor-specific instructions into a permanent claim.

What Rivul MCP returns

A useful search result needs enough structure to identify, compare, and open a work. Rivul MCP normalizes provider records into a bounded response instead of passing a provider-specific payload straight through to the caller.

• IdentityTitle, authors, publication year, venue, DOI, and stable record identifiers when available.
• Discovery contextWork type, citation count, topic, field, and a bounded abstract excerpt when available.
• Source locationsPublisher, repository, DOI, and open-access links when the underlying record supplies them.
• PaginationTotal matches, page size, offset, and whether another page is available.

Metadata availability varies by publication. A missing DOI or open-access URL should be treated as missing data, not inferred. The original publication is always the authority for the paper's text and findings.

Worked example: testing an agent search

After connecting Rivul MCP, ask the client: “Use Rivul MCP to search for Attention Is All You Need, then get the full record for the first result. Return the title, authors, year, venue, work type, DOI when available, topic, field, open-access status, and source URL.”Confirm that the client calls search_papers and then get_paper_details, because the search result is a short record and venue, work type, topic, field, and links come from the full record. Check that it preserves the source links. Then ask it to callusage_status so you can confirm that the connection is bound to the expected workspace and its current fair-use usage.

A literal query may rank books or chapters above journal articles. Refine the request when the research question needs a particular work type, date range, citation threshold, or open-access constraint. The model can help formulate that request, but the researcher still evaluates relevance and evidence quality.

The public contract and its boundaries

Rivul MCP documents what a customer needs to connect, search, inspect usage, and revoke access. Internal provider credentials, infrastructure addresses, credential formats, security controls, and operational thresholds are not part of the public integration contract. That separation keeps the documentation useful without turning implementation detail into a promise or an attack map.

Available capabilities and coming-soon work are labeled separately. Deterministic lookup by DOI, richer open-access resolution, related-paper discovery, and landscape reports are product direction until they are implemented. Read the Rivul MCP features page for the current customer-facing capability summary, then open MCP setup under Rivul MCP in the app for the exact setup guide supported by your client.