The BWG Global Research connector lets an AI assistant search and read BWG Global's primary research library on your behalf, inside the tool you already use. It is a remote Model Context Protocol server: you connect once, sign in with your BWG portal account, and the assistant can then search research, quote from it, and link you back to the portal to read the full document.
https://mcp.bwgglobal.com/mcpThe connector exposes two tools to the assistant.
bwg_search_documentsSearches the research library and returns matching documents.
| parameter | type | default | notes |
|---|---|---|---|
query |
string | "" |
Free text, matched against titles and research content. Empty returns the newest documents you are entitled to. |
days_back |
integer | 365 |
Only documents published in the last N days. |
top |
integer | 10 |
Maximum results, capped at 20. |
Each result carries the document's title, type, publication and event dates, companies and research leads covered, a link into the portal, and a short excerpt. Where your subscription includes them, results may also carry transcript excerpts, BWG sentiment ratings, and thematic ticker baskets.
Excerpts are previews, not the document. Research bodies run to tens of thousands of characters; the excerpt is capped at 2,000. The portal link is always the destination for the full text.
fetchReturns the full text of one document, a page at a time.
| parameter | type | default | notes |
|---|---|---|---|
document_id |
string | — | The id from a search result. |
part |
string | "research" |
"research" or "transcript". |
offset |
integer | 0 |
Character offset to start from. |
limit |
integer | 100,000 | Characters to return, capped at 100,000. |
Long documents are paginated — the response reports where the next page starts so the assistant can continue without you asking.
The connector uses OAuth 2.1 with PKCE and supports Dynamic Client Registration, so most MCP-capable clients can connect by URL alone.
https://mcp.bwgglobal.com/mcp.You are signing in as yourself. The connector never asks for, and never accepts, another user's identity.
Protocol revisions. The server speaks MCP revision 2026-07-28 on the
modern discovery path and negotiates 2025-11-25 and earlier for clients that
use the initialize handshake. Both are live; your client picks whichever it
supports, and no configuration is needed either way.
Results are scoped to your own entitlements, which come from your BWG subscription. Two people running the same search can legitimately get different results, and that is not an error.
Your subscription controls:
Access to the connector itself is a separate entitlement on your subscription. If it is not enabled, the assistant will tell you so and name your account manager.
The assistant may report one of two things:
| what you see | what it means | what to do |
|---|---|---|
"Your user is not entitled to the MCP Connector. Please reach out to <name>, your Account Manager…" |
Your subscription does not currently include connector access. | Contact the named account manager. |
| "Research access could not be verified right now… usually temporary." | A transient fault on our side. Your subscription is not in question. | Try again shortly. If it persists, email mcp@bwgglobal.com. |
An empty result means nothing matched within your entitlements — not that the research does not exist.
For anything else — connection failures, sign-in loops, results that look wrong — email mcp@bwgglobal.com with your assistant (Claude, ChatGPT, or other), roughly when it happened, and the query you ran.
What leaves BWG. When you use the connector, the search terms you (or the assistant on your behalf) send, and the research content returned — titles, metadata, excerpts, and any full text you fetch — are processed by whichever assistant you have connected. That assistant is operated by a third party, not by BWG Global, and their terms and privacy policy govern what they do with it. Connect only through a provider your firm permits.
What BWG records. Each tool call is logged for operational and security purposes: your contact identifier, the client application, the arguments sent, the identifiers of documents returned, timings, and the outcome. Document text is not logged.
What BWG reads about you. To decide what you may see, the connector reads your entitlements and contact record from BWG's CRM on each session, cached briefly. It does not read anything else.
Retention, lawful basis and your rights over this data are covered by the BWG Global privacy policy: https://bwgglobal.com/about/privacy-policy/
User prompt: "What's BWG's view on Nvidia — give me the full picture across research types."
What happens:
- The assistant searches by company name, then runs a second query using the ticker (e.g. NVDA) alongside "channel research," since Channel Research documents are often titled by ticker rather than company name.
- Results spanning Forum Synopsis, Channel Research, and Sector Pulse are returned together, so a handful of weekly forum mentions don't crowd out a single deeper quarterly report.
- Each result includes the document type, publication date, and portal link.
User prompt: "What did BWG publish today across all sectors?"
What happens:
- The assistant runs a broad query with days_back set to 1 rather than a topic-specific search, since same-day content can be missed if it hasn't yet been indexed under an expected keyword.
- All documents published that day are returned across sectors — Sector Pulse notes, Forum Synopses, Channel Research — with titles and links.
- If nothing has posted yet, the assistant reports that plainly rather than assuming the search failed.
User prompt: "Build a long/short basket for the AI data center buildout slowdown theme."
What happens: - The assistant searches thematic and Noonum-tagged strategy content. - It returns a single flat table — ticker, company name, exchange, conviction, long/short side, and a dated, linked source — with one row per ticker occurrence rather than silently deduplicating repeats across documents.
User prompt: "Pull the full text of that Channel Research report on Adidas you just found."
What happens:
- The assistant calls fetch with the document_id from the prior search result.
- It returns the research body starting at offset 0, and automatically requests the next page if the document runs longer than one response.
Please include your assistant, the approximate time, and what you asked. If you are reporting missing research, naming the document or company helps us tell an entitlement question from a search one.