CrawlCheck

Registry · Capabilities · machinerelations.ai

Capability registry

What machinerelations.ai declares an agent can do

Measured 2026-10-30 03:33 UTC from the declarations the site publishes (record yrq3v3titt). Each capability is shown as declared, under the policy class that decides whether anything may call it, beside what a call observed.

16declared
8public read-only
1called and verified
0declared ≠ observed
agent_skill, mcp, openapiprotocols
Public read-only: 8Authenticated read-only: 6Metadata: 2
Public read-only 8 · Authenticated read-only 6 · Metadata 2

Declared versus observed

CapabilityPolicy classDeclaredObserved
getIndex
openapi · GET · https://machinerelations.ai/api/mri/v2/
Public read-onlyHTTP GET with no declared security
auth: none · required inputs: 0
redirected (not followed): /api/mri/v2 HTTP 308 · application/json
listCategories
openapi · GET · https://machinerelations.ai/api/mri/v2/categories
Public read-onlyHTTP GET with no declared security
auth: none · required inputs: 0
behavior verified HTTP 200 · application/json
getCategory
openapi · GET · https://machinerelations.ai/api/mri/v2/categories/{category}
Public read-onlyHTTP GET with no declared security
auth: none · required inputs: 1
not attempted
needs 1 required input: nothing is invented to fill them
getCitedSources
openapi · GET · https://machinerelations.ai/api/mri/v2/categories/{category}/{questionShape}
Public read-onlyHTTP GET with no declared security
auth: none · required inputs: 2
not attempted
needs 2 required inputs: nothing is invented to fill them
getDomain
openapi · GET · https://machinerelations.ai/api/mri/v2/domains/{domain}
Public read-onlyHTTP GET with no declared security
auth: none · required inputs: 1
not attempted
needs 1 required input: nothing is invented to fill them
listEngines
openapi · GET · https://machinerelations.ai/api/mri/v2/engines
Public read-onlyHTTP GET with no declared security
auth: none · required inputs: 0
not attempted
outside the per-scan call budget (2 GETs, 1 MCP tool)
getEngine
openapi · GET · https://machinerelations.ai/api/mri/v2/engines/{engine}
Public read-onlyHTTP GET with no declared security
auth: none · required inputs: 1
not attempted
needs 1 required input: nothing is invented to fill them
getEngineCategory
openapi · GET · https://machinerelations.ai/api/mri/v2/engines/{engine}/categories/{category}
Public read-onlyHTTP GET with no declared security
auth: none · required inputs: 2
not attempted
needs 2 required inputs: nothing is invented to fill them
mri_list_categories
mcp · https://machinerelations.ai/mcp
Authenticated read-only
never called automatically
the tool declares readOnlyHint
auth: declared
not attempted
requires declared: never called without the operator's own credentials
mri_get_category
mcp · https://machinerelations.ai/mcp
Authenticated read-only
never called automatically
the tool declares readOnlyHint
auth: declared
not attempted
requires declared: never called without the operator's own credentials
mri_get_cited_sources
mcp · https://machinerelations.ai/mcp
Authenticated read-only
never called automatically
the tool declares readOnlyHint
auth: declared
not attempted
requires declared: never called without the operator's own credentials
mri_get_domain
mcp · https://machinerelations.ai/mcp
Authenticated read-only
never called automatically
the tool declares readOnlyHint
auth: declared
not attempted
requires declared: never called without the operator's own credentials
mri_list_engines
mcp · https://machinerelations.ai/mcp
Authenticated read-only
never called automatically
the tool declares readOnlyHint
auth: declared
not attempted
requires declared: never called without the operator's own credentials
mri_get_engine
mcp · https://machinerelations.ai/mcp
Authenticated read-only
never called automatically
the tool declares readOnlyHint
auth: declared
not attempted
requires declared: never called without the operator's own credentials
machine-relations
agent_skill · https://github.com/Bramwell-Inc/machinerelations-plugins/releases/download/v1.1.5/skill-machine-relations.tar.gz
Metadata
never called automatically
an instruction file an agent reads; not an operation
auth: none
not attempted
on another origin (github.com): only the scanned origin is called
machine-relations-index
agent_skill · https://github.com/Bramwell-Inc/machinerelations-plugins/releases/download/v1.1.5/skill-machine-relations-index.tar.gz
Metadata
never called automatically
an instruction file an agent reads; not an operation
auth: none
not attempted
on another origin (github.com): only the scanned origin is called

Where the declaration and the call disagree

None: every capability that was called answered as declared.

Signed certificate

This page’s rows, signed by the observer’s Ed25519 key: every capability’s id, endpoint, safety class, policy class and observed state, the declaration hashes and the record’s sealed manifest. The certificate is derived from the record alone, so the same record always yields the same id.

certificate_id ccap1:f17da05187eb5d14b40047fc167b309aaba7ed5257327e59ba7dfcf8fcba370c
alg Ed25519 · kid 1hHmE3fo1rg5Ig70wuFtJ5wBc7G-sUTgUJkxyPWFRps
sig Ni2aCfpIwK5k8lsR5jJAbd6wHkd0HxPF9TnLg-pqPB6DqxbBS6RSD83ffNLKPV9sp1q7sFYiQHGxX_unL3U9Cw

The certificate as JSON · verify it now · the key directory · machine-readability certificate

To verify it yourself: SHA-256 the certificate object as canonical JSON (keys sorted, no whitespace); the hex digest must equal certificate_sha256 and the id; then check the Ed25519 signature over "crawlcheck-capability-certificate-v1\n" followed by that hex digest, with the key in the directory whose RFC 7638 thumbprint is the kid.

Policy classes

Policy classCalled automatically?Rule
Public read-only
public_read_only
yes, to verifyMay be called without credentials. CrawlCheck calls it only to check the declaration: GET with no input, on the scanned origin.
Authenticated read-only
authenticated_read_only
neverNever called by CrawlCheck. An agent needs the user's own credentials and consent.
Reversible write
reversible_write
neverNever invoked automatically. Needs the owner's written authorisation and a test account.
Irreversible write
irreversible_write
neverNever invoked automatically. Deletes or changes something that cannot be undone.
Financial
financial
neverNever invoked automatically. Moves money or creates a charge.
External communication
external_communication
neverNever invoked automatically. Sends a message, email or notification to someone.
Identity and accounts
identity
neverNever invoked automatically. Creates, signs in to or changes an account or credential.
Unknown
unknown
neverNever invoked. Its effect cannot be read from the declaration, so it is treated as the riskiest case.
Metadata
metadata
neverNothing to invoke: a listing, a server card or a skill file, not an operation.