GeoComply announced the rollout of GeoComply MCP on 29 September 2026, connecting AI assistants to its location intelligence, risk signals and compliance data. According to the company announcement, fraud and compliance teams can investigate users, devices and transactions through the AI tools they already use, with supporting evidence returned alongside the analysis.
MCP stands for Model Context Protocol: a standard that lets an AI assistant access tools and data from another system. Here, it provides a connection to GeoComply’s investigation capabilities, giving analysts a way to ask questions without assembling each query themselves.
GeoComply had already promoted MCP in its 12 August masterclass programme. The September release marks a rollout announcement following that earlier public introduction.
Reading risk signals without changing accounts
The product documentation describes investigations into shared devices, connected accounts and suspected location spoofing, alongside compliance reports and explanations of failed checks. It also draws a clear boundary around those capabilities: MCP reads and reports information. It does not change account status or take case actions.
GeoComply says access follows the permissions already assigned to the analyst and organisation. Calls are recorded with attribution to the person or agent making them. An engineering team can connect its own agent, while analysts can work through compatible assistants or GeoComply’s own back office.
The launch release cautions that AI output can be incomplete or inaccurate and requires human review. The announcement provides no measured results from operators demonstrating faster investigations or better detection.
Where operators should test the handover
Our assessment is that the useful test begins with a real case. A shared device may connect several accounts, but the connection alone does not explain whether they belong to a household, a coordinated fraud operation or another situation. Reviewers need enough context to challenge an initial interpretation before deciding what follows.
A pilot could compare cases investigated through MCP with the team’s existing process, recording the time spent gathering evidence and checking the answer. Any time saved has to survive that second step. Teams could also test whether another reviewer can reconstruct the reasoning from the case record.
With regulators examining operator AI use, including the separate Massachusetts sportsbook AI review, that handover deserves attention. Connecting an assistant to risk data creates a new route to evidence; the operator still needs a clear owner for the decision.