Tutu has opened its services to AI agents via the Model Context Protocol (MCP), allowing external developers to integrate travel data into their applications. The protocol enables language models to access real-time information on flights, trains, buses, and hotels, bridging the gap between general advice and actionable booking steps.
In August, the company hosted an AI hackathon to explore potential use cases for this infrastructure. Out of 301 applicants, 150 participants across 71 teams were selected. By the end of the event, 42 teams submitted projects, with nine reaching the finals and three taking top prizes. Huawei served as a hardware partner for the prize fund.
Hackathon projects highlighted new user scenarios, such as group travel planning from different locations, discovery-based searches without fixed destinations, and budget-constrained trip optimization. These approaches demonstrate that user intent often exceeds the capabilities of traditional search forms.
Tutu emphasizes that while agents can handle search and comparison, critical actions like passenger data entry and payment remain on the company's website. This human-in-the-loop approach is a security measure against prompt injection attacks, which were observed twice during the hackathon.
The company views MCP as a new interaction channel rather than a replacement for its existing platforms. Future development focuses on balancing agent autonomy with strict control over tools and data to mitigate risks associated with automated transactions.
