What Is MCP for AI Research?

MCP in plain language

Model Context Protocol, or MCP, is an open standard for connecting an AI application to external tools and data. The AI client discovers what a server offers, reads each tool's input schema, and can call an appropriate tool during a conversation.

MCP is the connection layer. It does not make a language model more accurate by itself, and it does not replace source evaluation. The quality of a research answer still depends on the records returned, the papers opened, and the researcher's judgment.

• Read the official MCP server overview
• Read the official MCP tools specification

What you need before you start

Every Rivul AI account can use Rivul MCP at no cost. Rivul MCP is set up for a workspace the first time you open Rivul MCP in the app, so open it once before connecting a client; until then a correct client configuration still returns an authorization error.

Rivul MCP has no fixed daily or monthly quota. Calls are still counted so usage_status can report them, and fair-use limits per minute, hour, and day stop a runaway client. Treat usage_status as the authority for your own workspace.

• Read usage and fair-use limits
• See what is free

Set up the Rivul MCP server

Rivul MCP exposes one remote MCP server. Point your client at the endpoint below over Streamable HTTP; there is no local process to install and no separate host to configure.

Authenticate one of two ways. Browser authorization lets the client register itself, send you to Rivul to choose a single workspace, and approve bounded scopes without ever receiving your password. A dedicated workspace MCP key suits clients that cannot complete the browser flow; create one key per client, keep it out of tracked files and screenshots, and revoke it when the connection is no longer needed.

Rivul MCP requests two read-only scopes. search:read covers scholarly search and paper lookup; usage:read covers the workspace's usage and fair-use limits. A client granted neither can connect and still fail every call, so confirm scopes rather than assuming a completed consent screen is proof of access.

Endpoint   https://rivul.ai/mcp
Transport  Streamable HTTP
Auth       OAuth 2.1 browser authorization, or a workspace MCP key as a bearer token
Scopes     search:read, usage:read
Tools      search_papers, get_paper_details, usage_status (all read-only)

• Read the official MCP authorization specification
• Read the full tool reference

Choose your client

Rivul MCP works with Claude Code, OpenAI Codex, Cursor, Google Antigravity, Meta Muse Code, and any client that supports remote MCP servers. Each guide covers where to add the server, how to sign in, how to check the tools, and how to disconnect.

For a client without a guide here, start from its own remote HTTP MCP documentation and use https://rivul.ai/mcp as the server URL. Run one real tool call to confirm the connection before relying on it.

• Claude Code→
• OpenAI Codex→
• Cursor→
• Google Antigravity→
• Meta Muse Code→

A research question through MCP

Suppose a researcher asks an AI client: Find three highly cited papers about retrieval-augmented generation and return their DOI or source links.

1. Step 1: The client identifies search_papers as the relevant Rivul MCP tool.
2. Step 2: It sends a bounded query and result limit to Rivul MCP instead of inventing a bibliography from model memory.
3. Step 3: Rivul MCP returns short structured records with available identifiers and citation metadata. For source locations, the client calls get_paper_details with a result's openalex_id to get the full record, including open-access and landing links.
4. Step 4: The client summarizes the records for the researcher, who opens the original papers before using their findings as evidence.

The arguments below are what the client sends. Searches are free under fair-use rate limits. The tool reference lists every field a record can carry.

This flow improves provenance because the search result and its identifiers remain visible. It does not prove that a paper supports a later claim until the relevant source text is read and evaluated.

{
  "query": "retrieval-augmented generation",
  "limit": 3,
  "sort": "most_cited"
}

• See every parameter and response field

What Rivul MCP can and cannot do

Rivul MCP can search scholarly records, look up one record in full, return citation metadata and source locations, and report the approved workspace's usage state. Its current tools are read-only.

Rivul MCP does not read a paper automatically, guarantee that a result is relevant, validate a scientific claim, edit a Rivul AI manuscript, or provide unrestricted access to publisher full text. A metadata record is a discovery lead, not evidence for a finding.

Available clients, authentication paths, and allowances can change as the clients evolve.

Troubleshooting

The browser completes consent but the client never reconnects. The grant exists on the Rivul side while the client is still waiting for its callback. Disconnect the incomplete grant in Rivul MCP, then start login again rather than approving a second time.

The wrong workspace was approved. Authorization binds to exactly one workspace. Disconnect the grant in Rivul MCP and sign in again against the intended workspace. Never paste a session cookie or your Google password into a client to work around this.

Tools are discovered but never called. Discovery and invocation are separate steps, and a client can pass the first while failing the second. A prose answer with no tool invocation is not a passing test. Check your client's guide for a recorded runtime boundary before assuming a Rivul fault.

Calls fail after working earlier. Read usage_status, which never counts toward usage. A rate limit and a revoked credential produce different errors: a revoked key returns HTTP 401, while a rate limit asks the client to retry later.

• Connect Claude Code to Rivul MCP
• Connect OpenAI Codex to Rivul MCP
• Open the Rivul MCP page
• Open Rivul MCP