What Q4 MCP brings to your AI workflow

Q4 is a remote Model Context Protocol (MCP) server for Saudi listed-company research. Connect it once and the AI client your team already works in can answer questions about Tadawul issuers from filings, earnings call transcripts and financial data — with a citation attached to every figure it returns.

Why a general model cannot answer a Tadawul question

Ask a general-purpose language model for a Saudi issuer’s revenue line, a segment breakdown, or what management guided on the last earnings call, and you get one of two unusable answers: fluent prose with nothing behind it, or a refusal. In equity research, investor relations and corporate strategy, a number you cannot trace back to a filing is a number you cannot publish.

Q4 MCP closes that gap without moving your team into another tab. It is not a chatbot bolted onto a data vendor — it is a research layer your AI client queries directly, returning primary-source excerpts with citation metadata instead of unsourced summary.

What Q4 MCP actually provides

Coverage

Tadawul-listed issuers: company profiles, annual reports, investor presentations, quarterly filings, earnings calls, news and Argaam coverage — one Saudi research corpus rather than per-company PDF hunting.

Citations

Sourced results carry company, document, page, section, time range, quote and a Q4 app path where available, so any figure in an answer can be opened and verified at source.

Earnings calls

Call summaries, AI insights and speaker- and timestamp-level passages, with the full transcript available opt-in rather than dumped into context by default.

Filings

Filing metadata plus page- or query-scoped excerpts — the layer where units, year-over-year columns and neighbouring statement lines get confirmed before a number leaves the desk.

Transport

Streamable HTTP at api.q4.sa/mcp with OAuth 2.1 and PKCE, so the same connector works in Claude, ChatGPT Developer Mode and Cursor.

Four tools, one research loop

The connector exposes four tools, and they are deliberately split between discovery and audit. That split is what keeps an AI-assisted workflow defensible.

• q4_get_company — resolve an issuer by id, symbol or name and list what Q4 holds for it. Start here: an empty document list means no filing of that type and period exists, not that the search missed it.

• q4_search — ranked hits across PDFs, transcripts, news and Argaam, each with citation metadata. This is discovery, not the audit layer: snippet scale can disagree with the source excerpt, so nothing gets published from search alone.

• q4_get_document — filing metadata with page- or query-scoped excerpts. This is where a figure gets confirmed, including its units and the columns around it.

• q4_get_earnings_call — summary, insights and speaker or timestamp passages for a catalogued call, so a quote arrives with who said it and when.

Recommended loop: catalog, then search, then document or earnings call. Search finds the candidate; the document and call tools are what make the figure quotable.

Where the value shows up

• Equity research and buy-side analysts: move from “what did margins do across the last four quarters” to a cited excerpt per period, without opening four PDFs to find out.

• Investor relations teams: check how a peer worded guidance, or how your own last call reads back, with speaker and timestamp attached to the passage.

• Corporate development and strategy: build a first-pass view of a Saudi issuer — filings, calls, coverage — inside the assistant already open on your desktop.

• Engineering and data teams: query Saudi market research from Cursor while building the models, dashboards and internal tools that consume it.

The common thread is that the research and the verification happen in the same place. The analyst does not leave the assistant to go prove what it just said.

Every answer carries its audit trail

Citation metadata is the point of the connector, not a footnote on it. Because sourced results identify the company, the document, the page and section, the time range and the quote — plus a Q4 app path when one exists — a reviewer can open the exact passage a number came from. That is the difference between an AI answer you can circulate and one you have to re-do by hand.

Connect the client your team already uses

One production URL, one OAuth sign-in, and the same corpus in every client:

• Claude: add Q4 as a custom connector, leave additional request headers empty, and sign in with your Q4 account on the authorize page.

• ChatGPT Developer Mode: same URL, same OAuth sign-in.

• Cursor: can send a Bearer JWT from a Q4 login instead of running the browser OAuth flow.

• Health check, no auth required: api.q4.sa/mcp/health.

Security, limits, and what it does not do

Worth stating plainly, because procurement will ask and because an honest boundary is more useful than a vague one:

• Authentication is always required: OAuth 2.1 with PKCE (S256). Every tool call needs a valid subscriber token.

• Access tokens are 7-day JWTs. Q4 does not issue refresh tokens.

• Client registration is CIMD — the client ID is an HTTPS metadata URL. There is no dynamic client registration.

• Rate limit: 60 authenticated requests per user per minute. Tool timeout: 30 seconds.

• Internal errors are sanitized — database credentials and file paths are never exposed.

• There is no enterprise managed SSO and no per-organization document ACLs in this version: every entitled user sees the same Tadawul universe.

How to get access

A Q4 platform subscription already includes MCP access alongside the website. If you only want the connector, an MCP-only plan can be subscribed from the connector sign-in page, billed through Stream Pay — note that an MCP-only account does not include the Q4 dashboard. For entitlement questions, or to get a team set up, email contact@q4.sa.