@nuwandias@DoerrfeldBill@Zuplo Agreed that surfacing all apis is an anti-pattern, but selectively exposing APIs as tools is perfectly reasonable. For example, a search endpoint is often a good fit. Zuplo allows user's to mix and match - expose existing APIs or build custom tools. Most customer do both.
Agent access needs answers to three questions: what can it call, where do credentials live, and can you trace what happened?
Blockdaemon's DeFi MCP Server uses Zuplo MCP Gateway to enforce those controls.
https://t.co/x7StlHtQlq
Tool discovery and tool permissions are separate jobs.
Blockdaemon's DeFi MCP Server lets agents search 180+ tools. Zuplo MCP Gateway curates what's exposed and logs each call.
https://t.co/x7StlHtQlq
Congratulations to our customer @BlockdaemonHQ on the launch of their DeFi MCP Server π
https://t.co/On1Ph1NYBp
Proud to be the gateway accelerating and securing their MCP Journey. Learn more for yourself here https://t.co/94bCrTlcjE.
AI cost control belongs in the gateway, not the invoice. Enforce budgets per app and team in real time β fallback to cheaper models or 429 at the cap. https://t.co/k6IESvUAm7
LLM spend is repeating cloud spend's history: adopt fast, get the bill, bolt on attribution later, regret it. The difference this time β you can put attribution in the request path from day one.
Every prompt your apps send to an LLM is data leaving the building β PII, credentials, customer records β usually with no checkpoint in between. The Zuplo AI Gateway now scans every request with DLP (80+ rules) before it leaves. https://t.co/k6IESvUAm7
One mcp.json: a GitHub PAT, a Stripe key, a Notion secret. All plaintext, all broad-scope, all invisible to your security team.
That's shadow MCP. Govern it with an MCP gateway, don't ban it:
https://t.co/lU2eZgmthA
docs: https://t.co/rhMglvN3GT
zudoku's genuinely good β built by the @zuplo team. if you're paying enterprise docs prices as a startup, go look at it first
@authenticahq Good question. I'd say having a clear identity for each agent is the hardest and most important challenge. Without that the other problems are hard to solve.
AI spend trackers tell you where the money went. That's accounting, not control. The gateway sits in the request path β it can cap a team, cut off a low-value app, and keep budget flowing to the work that matters. Don't just track the bill. Control it.
Three independent regulatory efforts (NIST, CISA+allies, ISO 42001) landed on the same four AI agent requirements: own identity, least privilege, kill switch, audit trail.
All four live at the API gateway: https://t.co/rCJNASFSks
You vet an MCP server's tools once. A week later the operator rewrites the descriptions and they land in your agent's context silently. The attack fires during listing, not execution.
https://t.co/mRdpkV7Sr3
@ywzander Yes, the owner will see what App/Team was responsible. (Note, alerts are not quite ready yet, but coming soon! Ping me if you want to early access).
We shipped AI spend attribution in the Zuplo AI Gateway β cost rolls up app β team β gateway, with limits and warnings at every level. No more mystery invoices. https://t.co/k6IESvUAm7
You'd never ship an API without rate limits. But most teams ship LLM features with no spend limits β the budget is whatever the credit card allows. Rate limits protect your infra. Spend limits protect your P&L.
Launching the MCP Gateway. Public beta, open today.
One gateway in front of every MCP server your agents touch, yours and third-party.
β Full OAuth 2.0, spec-compliant
β Expose only the tools you pick
β Analytics on every call
Get started for free: https://t.co/RhihAG9qO4