Trading Central officially launched its Model Context Protocol server on September 21, 2026, after a six-month soft launch. The market research provider is based in Ottawa and Paris. The Trading Central MCP Server lets brokerages and financial institutions pull the firm’s proprietary, licensed research into AI agents and chatbots. It arrives in a structured, machine-ready format.

Trading Central has been producing research since 1999. It is registered with the SEC, the SFC and Anacofi, and holds ISO/IEC 27001:2022 certification.

Rights Clearance, Not Retrieval

Any competent engineering team can already wire a language model to a market data feed. Establishing who carries the liability when the resulting answer reaches a retail account holder is the harder part. No amount of engineering solves it.

That is the gap Trading Central is selling into. A regulated brokerage that lets a chatbot answer client questions about a security inherits responsibility for what the chatbot says. Suppose the model was grounded in scraped commentary, unvetted sources, or content the firm has no license to redistribute. Then the brokerage has manufactured a compliance problem. It will not surface until someone complains.

Chief executive Alain Pellier framed the launch as a data-quality point: an AI tool can be no more trustworthy than its inputs. The sharper framing is contractual. The company says every data point it serves is licensed and rights-cleared. That promise is what a compliance officer is buying.

Why MCP and Not Another API

Trading Central already distributes through iFrames, APIs and widgets. Adding MCP is less a technical necessity than an acknowledgement of how integration is now being done.

MCP standardizes how a model discovers and calls external tools, which removes the bespoke glue code each vendor integration used to require. Trading Central says its server exposes 51 tools across eight analytical domains, each returning structured JSON. Consider a brokerage agent that must check a price, pull a technical view and surface an earnings date at once. Three custom API integrations versus three MCP tools is a difference of weeks of engineering.

This is part of a visible pattern. Morningstar and PitchBook have announced MCP integrations, and Interactive Brokers has run its AI integration on MCP since launch. Fintechbits has tracked both in coverage of Morningstar and PitchBook’s Gemini MCP work and Interactive Brokers’ MCP tooling. Financial data vendors have worked out that if agents become the interface, the vendor easiest to call becomes the default source.

The Hallucination Claim Needs Reading Carefully

The release claims brokerages can deploy chatbots “free of hallucinations” through the server. That phrasing is doing more than the underlying technology supports.

Grounding a model in licensed, structured research substantially reduces fabrication, and structured tool output is far better than free-text retrieval. It does not stop the model misstating what a tool returned. Nor does it stop the model combining two correct figures into a wrong conclusion, or answering confidently outside the tool’s coverage. No MCP server changes that, because the failure happens after the data arrives.

What the server genuinely delivers is narrower and still valuable. When the model is wrong, the firm can show what it was given, by whom, under what license, and using which methodology. That is auditability, and in regulated distribution auditability is worth more than an accuracy claim nobody can verify.

What Trading Central Leaves Out

Six months of soft launch implies real deployments, and the release names none. It offers no client references, no usage figures and no pricing. The product’s entire value rests on being trusted by regulated firms. One named brokerage would have been the strongest evidence available.

The four stated differentiators are security and registration, copyright clearance, explainability, and coverage across the trading journey. Three of those are properties Trading Central already had before it built an MCP server. The genuinely new thing here is the delivery mechanism, and the company would be on firmer ground selling it as exactly that.

What to Watch

The signal to look for is a named brokerage running the Trading Central MCP Server in production and saying so. Vendors announce MCP servers easily. Regulated firms put client-facing agents into production slowly. The first one willing to be quoted will show whether compliance teams have accepted the licensed-grounding argument.

The second thing to watch is pricing structure. If Trading Central charges per tool call rather than per seat, it is betting that agent traffic will exceed human traffic. That would be the clearest statement yet about where AI-driven research distribution is heading.

Fintechbits covers AI in financial services, market data and brokerage technology. Nothing here constitutes financial or investment advice. All analysis represents the editorial views of Fintechbits.