Trust Tools

What trust-tools MCP server does

The trust-tools MCP server connects an MCP client to public trust and verification functions from Obelisk Gate. It covers website security posture, organization ratings, token authenticity, agent proof verification, transparency-ledger state, and public service status.

The server is suited to workflows that need evidence about an HTTPS site, organization, token, agent execution, or transparency history. Its verification tools report structured claims and validation results rather than exposing secrets or underlying task data.

Tools and capabilities

• scan_trust runs a public trust scan against an HTTPS URL. The result covers TLS and security-header posture and includes an Obelisk Rating. The scan is read-only and SSRF-guarded.
• verify_token checks whether a JSON Web Token was issued by Obelisk Gate using ES256 and the expected issuer. It reports the token type, subject, assurance, and principal, without returning secrets.
• get_org_rating retrieves an organization’s public Obelisk Rating using its slug. The response includes a score from 0 to 100, a trust band, and a trend.
• verify_agent_run_proof checks the structural integrity, agent binding, delegation continuity, and commitments in an action-scoped run proof. It does not reveal or infer raw task data.
• get_agent_proof retrieves an agent’s public proof vector by opaque proof ID. Results contain independent claims with freshness and scope; they are not a scalar trust score or a claim that the subject is non-human.
• transparency_head reads the current signed head of Obelisk’s tamper-evident receipt ledger. A caller can retain the head and compare it later to establish that the history only extended.
• gate_status reports public liveness facts, including the issuer, served code revision, OIDC availability, and supported protocols.
• explain_rating interprets an Obelisk Rating number by returning its trust band, describing the public scan’s scoring, and identifying factors that typically change the score. The result is deterministic by default; narrate:true adds a cached, one-line model-written narrative.

How it works

The trust-tools MCP server presents these operations as MCP tools. Several operations read public information, while the verification operations validate supplied tokens or proof material. explain_rating is a pure interpretation of the supplied number and does not read external data.

transparency_head provides the current ledger checkpoint, which can be pinned for a later comparison. get_agent_proof and verify_agent_run_proof serve different purposes: the former reads public claims by proof ID, while the latter validates the structure and continuity of a supplied run proof.

Limitations and notes

The listed operations are narrowly scoped to public ratings, public status, verification, and trust evidence. Token verification never returns secrets, and agent-proof operations do not expose or infer raw task data. An agent proof is not presented as a single trust score or evidence that an agent is non-human.

scan_trust accepts HTTPS URLs and is read-only. explain_rating does not calculate a new rating or fetch source data; it explains a number. The available material does not specify installation commands, transport details, authentication requirements, or environment variables.

The trust-tools MCP server should therefore be evaluated for evidence and verification workflows rather than general website crawling, secret management, or task-data inspection.