Agentic AI Reshapes Public Access (Image Credits: Images.fastcompany.com)
Boston – City technology leaders have developed a mediated interface to enable AI systems to interact reliably with public resources amid rising machine-driven demands on government portals.
Agentic AI Reshapes Public Access
Government websites now face a surge in automated traffic from AI agents querying databases and navigating services. Much of this activity remains benign, focused on search and data retrieval. However, risks loom large as unchecked agents could overload systems, submit false requests, or hoard limited resources.
Traditional interfaces faltered under this pressure, forcing AI tools to scrape pages or improvise based on outdated training. Boston officials recognized the need for structure. They opted against outright blocks or unguarded openness, choosing instead a controlled intermediary layer.
Model Context Protocol Emerges as Key Tool
Anthropic introduced the Model Context Protocol, or MCP, roughly a year ago to connect large language models with APIs and data systems. Boston CIO Santi Garces highlighted its potential in a recent discussion. “It provides a way for large language models to interface with the kinds of resources we have in government,” he explained, serving as an intermediary for secure access to transit updates or service requests.
MCP defines tools in plain language, mapping natural requests to precise programmatic calls. This setup curbs the randomness of AI responses, ensuring deterministic outcomes. Cities gain governance over interactions, fostering trust in agent-driven services.
From Prototype to Practical Use
Boston launched its first MCP application, Open Context, linked to the city’s open data portal. Students from Northeastern University’s AI for Impact program built the initial version in fall 2025. The effort integrated with the city’s AI Launchpad tool for employees, streamlining data analysis workflows.
Early tests revealed strengths in dataset discovery but limits with large-scale processing. Developers shifted computation to the portal’s query capabilities, boosting efficiency and accuracy. Users now pose questions like restaurant counts in Boston, triggering live SQL queries against current data – no more reliance on stale training or web guesses.
Security Gains and Broader Horizons
AI scraping already floods boston.gov, raising fears of fraudulent service grabs or resource scalping. MCP middleware enables monitoring, identity ties, and blocks on unauthorized access. It strikes a balance, allowing legitimate uses while fortifying defenses.
Officials view MCP as digital public infrastructure, replicable via open-source sharing. Good metadata underpins success, aiding AI comprehension. Expansion targets service requests, though high-stakes areas like civil liberties stay off-limits per city AI policies.
- Reliable data pulls reduce hallucinations.
- Cost savings through portal-based computation.
- Equity boost for non-English speakers or those with disabilities.
- Interoperable across AI models.
- Governed access prevents abuse.
Key Takeaways
- MCP turns chaotic AI browsing into structured, secure queries.
- Boston’s prototype proves low-risk starts with open data yield big gains.
- Future scalability hinges on easy discovery and vendor adoption.
Boston’s MCP initiative signals a pragmatic shift, turning AI’s disruptive force into a public good through deliberate design. Other cities can follow by prioritizing data governance and prototyping intermediaries. What steps should your local government take next? Share your thoughts in the comments.
