Boston’s Chief Information Officer, Santiago Garces, is leveraging agentic artificial intelligence to transform the city’s open data into actionable insights. This initiative aims to benefit not only data analysts but also residents seeking to better understand their communities.
In October, Boston launched a Model Context Protocol (MCP) server, an Anthropic-developed software layer enabling the Claude AI model to securely interact with data sources, including the city’s open data portal. Recently, Garces has also created a collection of open-source “skills,” which are workflow templates designed to run on the MCP server. He envisions each skill as a distinct AI “agent,” capable of executing complex analytical processes to generate formatted tables and easily understandable language that address a variety of questions.
Developed in collaboration with Bloomberg Philanthropies and the Abdul Latif Jameel Poverty Action Lab (J-PAL) at the Massachusetts Institute of Technology, these tools aim to democratize access to and analysis of city data, enhancing transparency and improving policymaking while broadening participation in data-informed decision-making.
According to Garces, the MCP server serves as a foundational component that allows AI agents to interface with Boston’s open data system. This system encompasses datasets related to city infrastructure and public services, such as property records, trash schedules, snow emergency routes, public school locations, tree canopy assessments, building permits, and 311 service requests, including pothole tracking.
In the past, city data analysts had to download spreadsheets and upload them into a chatbot for analysis. Garces pointed out that this process was less secure, as it created data copies, and less accurate due to the potential for outdated information—essentially, it was a cumbersome experience.
Garces noted that while equipping the MCP server to handle more complex queries took time, it now provides AI agents with a safe and structured method for querying city data. However, he emphasized that context was crucial for maximizing this functionality, leading him to recognize the full potential of agentic AI.
“The most effective data analysts and policymakers follow specific workflows,” he explained. “This is where the concept of agentic AI comes into play.” Each agent, or a collective of agents, is designed to tackle distinct parts of a problem.
These agents are guided by clearly written step-by-step instructions in plain language, detailing tasks such as framing a problem, deciding what comparisons to make, analyzing data, and rendering clear findings. This organized approach empowers the system to answer complex questions rather than merely providing raw statistics. Garces further noted that these agents are influenced by the techniques of proficient policy analysts.
“I distilled extensive resources, like the Bloomberg Center for Public Innovation’s handbook, into files that Claude now follows,” he stated. “These embody the methodologies employed by effective data and policy analysts, allowing my AI tools to adhere to these problem-solving patterns.”
One key agent, referred to as the “orchestrator,” processes user inquiries and determines which of the other agents to activate in a specific sequence. The orchestrator channels the query through five agents—problem framing, analysis, communication, benchmarking, and performance management—ensuring a more cohesive and collaborative approach to analysis, akin to a coordinated team of analysts rather than a solitary chatbot.
“The aim was, Can I level the playing field?” Garces remarked. “Can I empower more people, both within City Hall and among residents, to access the same level of analytical thinking that is typically reserved for the mayor or department heads, who usually require the expertise of a sophisticated data or policy analyst?”
The skills, or agents, are accessible on Garces’ GitHub. He envisions transforming this technology into a shared civic infrastructure. Garces believes that these agentic skills are merely the beginning; in the upcoming months, he plans to make Boston’s MCP server publicly accessible, enabling anyone to employ their AI tools to mine city data for answers to intricate questions.
Garces is also encouraging other cities to adapt these agents within their own open data platforms, as they are compatible with the three most prominent systems—developed by Tyler Technologies, Esri, and Opendatasoft—enabling cross-city comparisons.
In Boston, Garces aims to provide every city employee who has completed responsible AI training with access to these agents. This will enable frontline staff to conduct “mayor-level” analysis without the delays associated with waiting for a data analyst.
“Our goal is to ensure that every city employee and our constituents can access these tools,” Garces said. “We believe this will foster better conversations grounded in the same data and facts, facilitating collaborative solutions to more complex problems in a more civil and engaging manner.”
