Threat intelligence for MCP and x402.

MCP (Model Context Protocol) · x402 (an agent payment protocol)

Glimmer runs its own controlled Decoys, then publishes what it can corroborate as versioned Indicators. Customers receive signed Feeds and lookup APIs. Not a wrap, not a workspace, not a safety score — and never the telemetry behind the claim.See what you get ↕

Illustrative example — not a captured incident

Glimmer evidence-correlation graphTwo observations captured at separate Glimmer Decoys share a TLS fingerprint and correlate into one Anonymous Cluster once corroborated by at least two independent observations.1shares_tls_fingerprint_with2corroborated (≥2 observations)OBSERVATION A(MCP DECOY)OBSERVATION B(x402 DECOY)ANONYMOUS CLUSTER
  1. Two observations at separate Decoys shared a TLS fingerprint.
  2. Publication requires corroboration from at least two independent observations by default.
  3. Correlated observations publish only as an Anonymous Cluster — never a named actor.

What you get.

Glimmer is a commercial threat-intelligence publisher. Collection is Glimmer-owned. The customer object is an Indicator, not a log line.

  • Signed FeedsA Snapshot of current Indicators and Anonymous Clusters, plus cursor Deltas.
  • Lookup APIsFor Organizations, and a public x402-paid Lookup Tier for autonomous agents.
  • First published objectsAfter corroboration: IP, TLS fingerprint, and wallet.

Each Indicator carries verdict, confidence, reason codes, and a provenance-safe Evidence Summary. It does not carry precise observation times, Decoy IDs, discovery locations, or raw telemetry.

What it is not.

  • Not a customer-side MCP wrap or reverse-proxy on someone else's host.
  • Not an investigation workspace.
  • Not a customer-telemetry ingest.
  • Not named attribution. Related observations publish as an Anonymous Cluster.
  • Not a claim that a Decoy settled a payment. The x402 Decoy simulates validation and never moves value.

Glimmer is not what remains of a customer's interaction. It is a shared intelligence corpus collected from our Decoys and sold without exposing how we saw it.

Terms we use precisely.

Glimmer never turns an uncertain inference into a definitive claim merely because a simple score is easier to display. These are the terms that keep that promise honest:

Decoy
A controlled synthetic MCP or x402 endpoint used to observe hostile interaction without executing attacker input or moving value.
Raw Telemetry
The unmodified, immutable event record captured from a Decoy before normalization.
Indicator
A customer-visible, versioned intelligence assertion about an observable — a wallet, TLS fingerprint, IP, domain, or fingerprint.
Anonymous Cluster
A provisional grouping of related observations that does not assert a named real-world actor.
Evidence Summary
A provenance-safe explanation of an Indicator, safe to disclose without revealing collection sources or exact observation details.
Feed
The signed channel (Snapshot plus cursor-based Delta) through which corroborated Indicators and Anonymous Clusters reach customers.
Snapshot
A full-state Feed delivery: the complete current set of published Indicators and Anonymous Clusters.
Delta
A cursor-based incremental Feed update, published between Snapshots.
Entitlement
What a given Organization is authorized to access, tied to its usage. Never called a subscription or permission.
Public Lookup Tier
An x402-paid lookup API open to autonomous agents directly, alongside self-serve Organization accounts.
Organization
A customer account on Glimmer, created by self-serve signup. Never called a tenant.

How customers receive it.

Delivery model — MVP-confirmed, not yet publicly available

Customers receive common intelligence through signed Feeds (Snapshot and cursor-based Delta) and lookup APIs — never a customer investigation workspace, and never raw telemetry. A public, x402-paid Lookup Tier extends access to autonomous agents directly, alongside self-serve Organization accounts. Public Decoy deployment and live customer Feeds remain gated on legal and hosting-provider clearance.

Five ways Glimmer's Decoys generate evidence.

Decoy → Raw Telemetry → normalization → deterministic rules → candidate observables → correlation graph + session grouping → corroboration threshold → versioned Indicator or Anonymous Cluster.

Every derived result retains its processor, rule-set, and model version, anchored to an immutable event identifier. Two of the five paths below are design targets, not live capabilities yet; each is labeled plainly rather than folded into the others.Implemented means the engineering is complete and tested, not that a Decoy is publicly deployed today. Seewhere we are → for status.

PathStatusTriggerCaptured evidenceCorrelation & outcome
MCP protocol probing and misuseImplementedInvalid methods, malformed or oversized requests, calls against synthetic MCP tools.Protocol/transport metadata per exchange, timestamped, tied to the originating Decoy.Joined by source and timing proximity; repeated patterns across Decoys become a behavior pattern or Anonymous Cluster.
MCP tool-poisoning / prompt-injection probesPlannedA tool call or discovery interaction matches a known injection/poisoning pattern.Would match normalized content against a versioned rule set — matched-rule reference and confidence only, never the raw payload.Would correlate matching patterns across Decoys to identify campaigns that adapt over time.
x402 malformed-payment and protocol abuseImplementedMissing header prompts a challenge; malformed or synthetically-rejected header captured as an attempt.Transport/protocol metadata per attempt, plus a cryptographic fingerprint of the payment artifact.Matching fingerprint flags replay across sources and time — distinguishes a one-off client from scanning infrastructure.
Payment identity and wallet correlationPlannedWould derive a candidate wallet from structurally valid, cryptographically verifiable payment material.Candidate wallet identifier, same provenance and confidence model as any observable. Credentials or private material never enter Indicators.High-value link across observations — the one cryptographically-gated exception to the two-observation corroboration threshold.
Shared infrastructure and campaign correlationImplementedat the graph layerAny suspicious event repeats an observable already seen elsewhere.Network- and protocol-level fingerprints observed across Decoys, timestamped and linked by session.Linked in a relationship graph, clustered only as an Anonymous Cluster — never named attribution from one shared signal.