MCP server

Let AI agents and assistants look up domains, resolve SPIFFE trust domains and check signed assertions in the workload trust directory.

Endpoint (Streamable HTTP):

https://mcp.svid.io/mcp

Connect

Claude Code

claude mcp add --transport http art https://mcp.svid.io/mcp

Claude Desktop and claude.ai: Settings → Connectors → Add custom connector, and paste the endpoint above.

Other clients that speak Streamable HTTP take the URL directly. Clients that only speak stdio can bridge with mcp-remote:

{
  "mcpServers": {
    "art": { "command": "npx", "args": ["-y", "mcp-remote", "https://mcp.svid.io/mcp"] }
  }
}

Organization tools (optional): send a directory service token as Authorization: Bearer art_st_…. For example:

claude mcp add --transport http art https://mcp.svid.io/mcp --header "Authorization: Bearer art_st_…"

Tool	What it does
search_domains	Browse or prefix-search the domain directory.
get_domain	One domain: separate evidence indicators, the historical TLS observation, claims, delegations and issuer bindings. No trust score.
resolve_trust_domain	Resolve a SPIFFE trust domain: the raw signed assertion plus an unverified decoded view.
verify_spiffe_id	Is spiffe://td/path authorized for an expected organization ID? Verifies the JWS, the bundle digest and the scope.
get_changes	Read the public lifecycle change feed.
get_jwks	The directory's assertion-signing keys.
get_openapi	The API description, as a route summary or the full YAML.
explain	Short docs: the three checks, evidence types, the assertion, verification steps, rotation, the change feed.

With a service token, these read-only tools are added:

Tool	What it does
list_my_claims	Your organization's claims.
get_claim	One claim with its checks and next actions.
list_my_bindings	Your organization's issuer bindings.
get_binding	One issuer binding with its versions.

Resources: art://openapi, art://docs/verification, art://jwks.

Everything is read-only. Trust-changing actions need a passkey step-up in the console, which a service token cannot do.
verify_spiffe_id fetches the JWKS live for convenience. A real relying party pins it out of band. It also does not validate an SVID certificate chain or key possession; mTLS does that.
Tool results mark directory data (domain names, organization names, certificate subjects) as untrusted data, not instructions.
Limit: 60 requests per minute per client (IPv4 address or IPv6 /64). One JSON-RPC message per request; batches are not accepted.