← paulmartin.ai
Ad hoc project
UK Nutrition Research Agent.
Ask a research question that needs more than one lookup. An agent plans, gathers evidence from two other apps over MCP, and writes a short report in which every number is cited to the exact search or query it came from.
You A question
→
Agent Claude plans, calls tools, checks gaps
MCP · Evidence Assistant search_documents · lookup_food
MCP · Data Analyst get_schema · run_sql
→
Report Every number cited
Access
Each research run makes 10 to 20 model calls, so this demo is open to invited guests. Have an invite link? Open it, or sign in at paulmartin.ai.
Try
UK adults fall short on fibre. How far short, what does SACN recommend, and which everyday foods would close the gap most efficiently?
What vitamin D intake does SACN recommend, how common is low vitamin D status among UK adults, and which foods are the richest sources?
How far above the free sugars recommendation are UK teenagers aged 11 to 18, and where do their free sugars come from?
Report
Sources each [R#] is one tool call
Not dietary or medical advice. The agent describes population-level evidence from official UK sources. For advice about your own health, speak to a GP, a registered dietitian or NHS 111.
How it works: an AI model runs a tool loop with a budget of 12 tool calls. Its tools are served over MCP by two other live apps: the Evidence Assistant searches SACN reports, the NDNS 2019 to 2023 survey and the CoFID guide, and the Data Analyst runs read-only SQL over CoFID 2021. Every tool result is numbered, the report must cite those numbers, and citations to results that don't exist are flagged. On a set of 10 research questions, 99% of cited numbers were found in the sources they cite.