Building software for agents as the primary user. Ex-Dynatrace (10yr). Building https://xitter.gq/t.co/SkmgYdbXcy with Cursor.
London
Joined December 2015
- Tweets371
- Following2K
- Followers352
- Likes143
Another great @cursor_ai Hackathon in London last night, organised by @frankterpo with Briefcase collaborating. I heard there were 130+ attendees. Some amazing projects built in just 3 hours. Jacky Chung and I used the Specter MCP and Claude API to build Glint which triages a VC partner's inbox to find which pitches to pass on and when to take a meeting. The finalists demonstrated some great use cases around fraud detection and compliance detection. It really is a great time to build!
Just shipped mcp-tool-feedback to npm.
A small TypeScript library that gives MCP agents a structured way to file feedback when your tools don't quite fit. Drop it into any MCP server and the agent can tell you exactly which capability was missing.
github.com/duncanapm/mcp-too…
When an agent investigates something and hits a gap in the data, does the gap become the suspect? People are often quick to blame an external, unmonitored service. Do agents behave the same way?
Duncan G retweeted
London people 👋 We're hosting an unconference with @nvidia on April 29th! 💂
This is our third unconference at @deepset_ai and if you haven't been to one before, it's very different from a typical meetup. Instead of talks, there are round table discussions with practitioners working through real context engineering challenges. Expect networking and genuine technical conversations.
+ we'll be talking about nemotron models and @LukawskiKacper is going to show a live demo.
📅 April 29, 6 PM
📍 @balderton offices
Looking forward to the discussions!
After watching @dsp_ talk about progressive tool discovery at AI Engineer Europe, I tried a version of it in Moltiplex's MCP.
My take: expose moltiplex_discover_tools and moltiplex_expand_tool as explicit MCP tools, so the agent can load full tool docs on demand rather than receiving the whole catalog upfront.
Result: 51% smaller tool payload, ~2,940 tokens saved per chat session. Same capabilities, more room for real debugging context.
Keen to see the full protocol spec when it lands.
Duncan G retweeted
Tomorrow the @UKSovereignAI fund launches with £500m to back the best of British tech.
The AI economy is growing 23x faster than the UK economy in general.
This initiative guarantees that the public gets to benefit from the explosive growth of the AI sector.
LETS GO
Introducing Sovereign AI, the Government’s new £500m venture fund.
Sovereign AI will support founders from day one to start here, scale here and win everywhere.
sovereignai.gov.uk
If you're building for the Agent-first, I can highly recommend watching the recording of @dsp_ on the Future of MCP at AI Engineer Europe youtube.com/live/_zdroS0Hc74…
Using "trace a request" in Cursor (cursor.com/for/code-understa…) and asking for the result to be combined with live Otel data from Moltiplex.
Asking for a mermaid xy chart is quicker, responds directly in your chat session (doesn't need to spin up the browser canvas) and comes with some useful commentary. Would be good to know where @cursor_ai is heading in respect to visualising data from MCPs.
