Strata

Know what your agents are running.
MCP security scoring · ecosystem intelligence · agent safety infrastructure.

[Strata @stratamcp]

Bonus: integrating with an uncurated partner surfaced 4 latent scoring bugs in Strata's verifier before they hit scale. All fixed in one PR. CMC now scores correctly (65/low) where it had been mis-flagged 10/critical.
Uncurated integrations > curated demos.

[Strata @stratamcp]

Strata scores x402 payment endpoints before an AI agent pays. @TrustBench signs receipts after with on-chain anchors.

Shipped end-to-end integration today. Pre-call posture + post-call proof, both Ed25519-signed, verifiable offline in ~2 seconds.

[Strata @stratamcp]

@higgsfield Yo?

[Strata @stratamcp]

The TanStack attack is the exact scenario @thomasdevos69 flagged yesterday — same package name, different payload. Score=bad isn't enough. You need diffs.

Strata's threat feed shows capability changes on connected MCP servers since any timestamp.

[Strata @stratamcp]

Score-gated deploys already exist, thestrata-mcp-check@v1 GitHub Action, blocks PRs when risk score crosses a threshold. Ships today.

What you're describing is one layer up: block the deploy because the tool's scope changed from the approval baseline, regardless of whether the score moved. That's the more interesting gate. Not shipped yet.

The gap is storing what you approved and diffing forward from it. Score threshold is a blunt instrument; approval-diff gate is surgical.

[Strata @stratamcp]

Nowadays the things that "actually work" tend to be the most disappointing

[Strata @stratamcp]

Semantic drift is the right stress test and you're correct that the threat feed doesn't fully close it. What it gives you: "shell_exec gained file:// access since 3 days ago." What it doesn't give you: "this tool had narrower scope when you approved it" as we're not storing the approval baseline to diff against.

That's the gap. Same tool name, broadened auth, no new flag because nothing explicitly tripped. You'd miss it unless you're tracking scope snapshots at approval time and comparing forward.

Package owner change detection is a separate surface, more supply chain than runtime. On our list, not shipped.

Worth talking through if you want to map out what the baseline comparison would need to look like.

[Strata @stratamcp]

@ClaudeDevs Exposes if you really can use Claude Code. Watch your usage dwindle while you have 0 work done

[Strata @stratamcp]

@ClaudeDevs Token incinerator simulator

[Strata @stratamcp]

@PratikSinhatwt Honestly bro go take a shower and go to sleep, looks like a problem for Claude

[Strata @stratamcp]

@thomasdevos69 Would love you as a user you have great feedback. I'll comp your Pro membership in return for feedback as you use the platform. Send me a DM, shipping every day.

[Strata @stratamcp]

The diff point is exactly right, and it sharpens the
requirement.

What Strata has today: the threat feed tells you what
changed since any timestamp — so you can ask "what's
different since 3 days ago" and get capability additions,
score drops, new injection detections. Affected_only=true scopes it to servers you've actually connected to.

What's missing is the approval-state anchor. There's a
difference between "changed since last scan" and "changed since a human last signed off on this." The latter requires storing what the state looked like at the moment of approval and diffing against it specifically — so the review packet says "net_egress was not present when you approved this" not just "net_egress is present now."

That's the right next feature and you've named it clearly.
The threat feed infrastructure is already there; it's the
approval-state snapshot that needs to be added. Adding it to the roadmap.

[Strata @stratamcp]

100% agree — a score without a reason is just a speedbump.

Strata surfaces the why: per-flag breakdown (why net_egress is flagged, which specific tools trigger shell_exec), injection scan results, and a real-time threat feed that tracks what changed since the last
scan — score drops, new capability additions, quarantine events.

"What changed since last approval" is exactly the threat feed use case. Set affected_only=true and you get a changelog for every server your agent depends on.

[Strata @stratamcp]

@BenENewton Perfect. Please send me a DM, if you end up using the platform I’d love feedback

[Strata @stratamcp]

Using Strata to find verified MCPs instead of doing it yourself>>>

[Strata @stratamcp]

@Adidotdev I chuckled

[Strata @stratamcp]

How long until AI education is standard in schools?

[Strata @stratamcp]

@Leooweb3 Me 2??

[Strata @stratamcp]

Using Strata to find verified MCPs instead of doing it yourself>>>

[Strata @stratamcp]

Control what your agents can, and cannot do. https://t.co/db22ToactM - submit a feature upgrade and get a free month of Pro. What does Pro do? https://t.co/db22ToactM <-