GSA schedules October 19 kickoff for government AI hackathon

GSA’s virtual Model Context Protocol Server and AI Agent Government Hackathon lists its kickoff for October 19, 2026, at 1 p.m. Eastern time, subject to schedule changes. Registered participants will receive access instructions; the official event page provides no public watch link. The assignment: build experimental connections between AI applications and government data or services.

A virtual kickoff, with registration already closed

The event is organized by GSA’s AI Community of Practice. Eligibility covers government employees with a .gov or .mil email address, including federal, state, local, territorial and tribal workers. Registration is closed, so the upcoming ceremony is not an invitation for new entrants. GSA says successfully registered participants will receive information on how to participate.

The calendar needs care: the page’s overview describes a September–November 2026 event, while its detailed schedule places the kickoff in October and the closing ceremony on December 16 at 1 p.m. Eastern. GSA explicitly warns that exact dates are still being finalized and may change. A halfway check-in is listed for November 3 at 12:30 p.m. Eastern.

Three ways to connect an agent

Model Context Protocol, or MCP, is an open-source standard for connecting AI applications to outside systems. Here, an MCP server would act as a bridge between an AI chatbot or assistant and the software interfaces that expose government information or services. GSA wants agency experts and developers to build those bridges around specific datasets and service tasks.

• Dataset access: let an AI application query an open government dataset.
• Service read: let an AI application retrieve information from a government service.
• Service write: let an AI application submit information to a government service inside a sandbox.

GSA’s intended output is an internal library of server prototypes. It says these could help AI systems answer plain-language questions, find relevant datasets and support analysis for decision-makers. Broader public access through AI interfaces is described as a possible future benefit, not a service being released at the hackathon.

Testing counts alongside the demo

Participants can work alone or in teams; GSA recommends no more than five people per team, combining software engineers with domain experts. Employees who understand their agency’s data are explicitly part of the intended workforce, not just programmers. Their suggested role includes identifying useful problems, explaining existing workflows and turning business needs into developer requirements.

Databricks, OpenAI and IBM are listed as industry partners. GSA promises training, technical office hours and mentorship during a six-week development window. Its estimated workload totals 40 hours: 10 for researching software interfaces and setting up tools, 10 for the initial build, and 20 for evaluation and improvements.

The judging criteria make testing a concrete requirement. Entrants must document their evaluation setup, inputs and results. Judges will look for accurate responses, graceful handling of errors and stable performance across test cases. They will also assess whether tools represent meaningful tasks for an agent, rather than simply copying the structure of the underlying software interface.

Final submissions must include documented code in a GitHub repository, a presentation and evaluation documentation covering testing methods, performance metrics, security considerations and lessons learned. Innovation, mission alignment and presentation also count toward judging. Winners will receive recognition from senior government officials; GSA says federal legal requirements prevent it from offering cross-agency monetary prizes.