Tweet fijado
MCP defines how AI agents connect to tools. It says nothing about how they discover which tools exist.
So I built a fix.
github.com/mariothomas/mc…
English
Mario Thomas
2.7K posts
Mario Thomas
@mariothomas
Head of Applied AI & Emerging Tech Strategy @AWSCloud | Chartered Director & Fellow @The_IoD | AI, Cloud, & Data Leader | @LSEnews Alumnus | Opinions my own
Llantwit Major, Wales Se unió Şubat 2009
2.4K Siguiendo4.1K Seguidores
Late last night a US government directive pulled Anthropic's Fable 5 and Mythos 5 frontier models offline for every customer, with no notice and no appeal.
Last year I set out the AI Sovereignty Trilemma: sovereign control, frontier capability, economical compute. You can hold two. Not three.
Most Boards quietly chose the last two and called it prudence. The bill for the third just arrived.
Anthropic@AnthropicAI
The US government, citing national security authorities, has issued an export control directive to suspend all access to Fable 5 and Mythos 5 by any foreign national, whether inside or outside the United States, including foreign national Anthropic employees. The net effect of this order is that we must abruptly disable Fable 5 and Mythos 5 for all our customers to ensure compliance. Access to all other Claude models is not affected. We apologize for this disruption to our customers. We believe this is a misunderstanding and are working to restore access as soon as possible. Read our full statement: anthropic.com/news/fable-myt…
English
“Is it a little bit homophobic to focus on the straights of Hormuz rather than the gays of Hormuz?”
No Kings protester, completely serious: “Yes, absolutely, I agree.”
English
This can't come soon enough here in the UK. As my parents get older, I'm really worried they'll lose the confidence to drive. Mum's already stopped, and Dad (ex-trucker) hates showing any weakness there. FSD would let them keep their independence, visiting friends, doing the shopping, getting around all without giving up their agency.
It's the same reason I'm excited for Optimus. The quicker we get there, the sooner one could give them home help, allowing them to live independently as long as possible.
English
Europe will love Tesla self-driving!
Due to the extreme regulatory burden of the EU, which in general stifles innovation in Europe, Tesla owners there have been stuck with basic lane-following.
Sawyer Merritt@SawyerMerritt
German TV reporter testing @Tesla FSD (Supervised) V14 in the country as public transport in rural areas: "I was genuinely impressed. In the situations where we experienced the system, it worked perfectly and safely. I hadn't expected that. Even in the bad weather conditions in the Eifel region. In many cases, it reacted at least as well as a human driver, if not better. If Tesla is ever allowed to roll out this system nationwide in Germany, I think it will have a major impact on mobility. And that will only be the beginning of some very significant changes in transportation..."
English
Mario Thomas retuiteado
For those of you interested in jury trials, this is Parliament at its best. 👇
English
@heyblake MCP defines how AI agents connect to tools. It says nothing about how they discover which tools exist.
So I built a fix.
github.com/mariothomas/mc…
English
The reference implementation is live right now at mcp.mariothomas.com
Three servers: articles and locations are public, documents is authenticated — demonstrating the governance pattern in practice.
Verify the DNS record yourself: dig TXT _mcp.mariothomas.com +short
Full source, spec, and deployment guide at the repo and a handy client guide to test it is here: github.com/mariothomas/mc…
English
MCP defines how AI agents connect to tools. It says nothing about how they discover which tools exist.
So I built a fix.
github.com/mariothomas/mc…
English
Every registry change is a pull request.
Attributed. Reviewed. Revertible. The read path is fully serverless — CloudFront and Lambda@Edge.
Governance lives in the write path.
That produces something boards and regulators will ask for: a queryable log of every agent that accessed the registry, what it requested, and when.
English
The record looks like this:
_mcp.yourdomain.com IN TXT "v=mcp1;registry=mcp.yourdomain.com/registry;publi…"
The registry it points to is itself an MCP server. Agents discover it using the same tools/list call they already make. Zero new client behaviour required.
Public servers return to any agent. Private servers surface only to authenticated agents. Same infrastructure, deliberate separation.
English
DNS has solved this before.
Email clients find mail servers via MX records. Services advertise endpoints via SRV records. _dmarc handles email authentication.
The same pattern works for MCP. One TXT record at _mcp.yourdomain.com points any compliant agent to your entire MCP ecosystem.
The full architecture is in the whitepaper: mariothomas.com/whitepapers
English
Without a discovery layer, every agent has to be manually told where every server lives.
10 agents. 20 servers. That's 200 configuration decisions — hard-coded at build time, maintained by hand, with no audit trail and no way for an agent to discover what it wasn't told about when it was built.
This is the n×m problem. It doesn't scale.
English
@AnthropicAI This is exactly what I wrote about on my blog last week. AI amplifies existing capability rather than replacing it — and may be dismantling the apprenticeship model that creates senior developers in the first place.
The verification premium is real. x.com/mariothomas/st…
Mario Thomas@mariothomas
English
AI can make work faster, but a fear is that relying on it may make it harder to learn new skills on the job.
We ran an experiment with software engineers to learn more. Coding with AI led to a decrease in mastery—but this depended on how people used it.
anthropic.com/research/AI-as…
English