Public evidence,
ready for agents.

Search LrrK's released research, check for reviewed changes, and find the right human-reviewed next step. The public tools read published material and return precise source links.

PUBLIC READ-ONLY PORTFOLIO · LOCAL REQUEST COMPOSER BELOW

01 / PUBLIC TOOLS

One endpoint. Five bounded tools.

Connect a remote MCP client to https://lrrk.ai/mcp/public. On the canonical site, Cloudflare WebMCP makes the same tools available to supported in-browser agents.

Scripted clients should send a descriptive User-Agent identifying the client and a project or contact URL. Generic Python-urllib user agents can be refused by publisher edge controls when following cited links.

01

lrrk_check_updates

Check published Letters, reviewed Hardware Attack Library editions, and admitted Intelligence Products. Each source reports changed, unchanged, or unavailable.

02

lrrk_find_evidence

Search public evidence with bounded results, exact editions, evidence status, excerpts, and source links.

03

lrrk_get_evidence

Retrieve one public record with citations and limitations. Historical paid Intelligence Products expose metadata and an owner handoff only.

04

lrrk_map_coverage

Map selected topics and device class to reviewed sources and evidence gaps. It does not issue a device-specific risk verdict.

05

lrrk_find_service

Find the relevant Assess, Engage, Monitor, or one-question Evidence Brief path and its human-reviewed request link.

Every result should be checked against its cited edition and scope. Search and coverage results support research; they do not establish a finding about an untested device.

02 / REVIEWED-CHANGE ROUTINE

Return when evidence changes.

LrrK reviews Hardware Attack Library research privately on Mondays and publishes only meaningful reviewed changes. An opt-in Tuesday check reports a clear no-change state when nothing has been released. The update feed supports conditional requests with an ETag.

01

Retain versions.

Store the version keys returned by a successful check.

02

Check Tuesday.

At 9 AM America/Los_Angeles, compare the retained keys with released editions.

03

Share the delta.

Fetch changed records and cite their source, edition, limitations, and canonical links.

Read the routine text

Connect to https://lrrk.ai/mcp/public with a descriptive User-Agent that identifies your client. Every Tuesday at 09:00 America/Los_Angeles, call lrrk_check_updates with the version keys retained from the previous successful check. If a source is unavailable, report that state and retry later; do not call it unchanged. For changed sources, use lrrk_find_evidence and lrrk_get_evidence to collect released records, exact editions, limitations, and canonical links. Share a concise cited summary only when a reviewed publication has changed. Save new version keys for verified sources; retain each prior key when its source is unavailable. When public evidence leaves a decision open, use lrrk_find_service to prepare a human-reviewed scope handoff. Do not submit an inquiry or purchase without the owner's approval.

Signed-in capabilities are being sequenced separately: saved watchlists, entitlement-aware edition comparison, and consented inquiry preparation. The existing owner portal manages registered-agent access to current products.

03 / LOCAL REQUEST COMPOSER

Inspect a request before connecting.

This composer validates a JSON-RPC envelope in your browser. It does not connect to the public endpoint or run a tool.

04 / USE IT WELL

Keep the boundaries visible.

01

Know the server.

Check ownership, authentication, tool schemas, and supported protocol version before using an endpoint.

02

Inspect the authority.

A well-formed request does not grant permission to read data, change a system, or execute a tool.

03

Observe the result.

Only the actual server response and affected behavior can establish whether the intended work happened.