MCP conformance rules

Catalog for :mod:app.mcp_conformance. Every rule cites an MCP specification reference. Blocking rules include CLX-4.3 transparency fields. Fetch via GET /v1/mcp/conformance/rules.

protocol.declared-capability-empty

• Category: protocol
• Severity: info
• Spec version: 2025-06-18
• Spec reference: https://modelcontextprotocol.io/specification/2025-06-18/basic/lifecycle
• Rationale: A declared capability that lists nothing gives an agent a dead end.
• Requires transcript: False

protocol.empty-page-with-next-cursor

• Category: protocol
• Severity: warning
• Spec version: 2025-06-18
• Spec reference: https://modelcontextprotocol.io/specification/2025-06-18/server/utilities/pagination
• Rationale: An empty page that still advertises nextCursor wastes a round trip per page.
• Requires transcript: True

protocol.error-code-non-standard

• Category: protocol
• Severity: warning
• Spec version: 2025-06-18
• Spec reference: https://modelcontextprotocol.io/specification/2025-06-18/basic/transports
• Rationale: An error code inside the JSON-RPC reserved band must be a defined code.
• Requires transcript: True

protocol.list-result-missing-items

• Category: protocol
• Severity: error
• Spec version: 2025-06-18
• Spec reference: https://modelcontextprotocol.io/specification/2025-06-18/server/utilities/pagination
• Rationale: A list result MUST carry its item array, even when empty.
• Requires transcript: True
• Reference: https://modelcontextprotocol.io/specification/2025-06-18/basic/transports
• Remediation: Include the method's items key (tools, resources, prompts, …) as an array.
• False-positive guidance: Skipped (never failed) when no transcript was captured.
• Fixture: mcp/unsafe/conformance/protocol-list-result-missing-items
• Scan modes: protocol, requires_transcript

protocol.missing-protocol-version

• Category: protocol
• Severity: error
• Spec version: 2025-06-18
• Spec reference: https://modelcontextprotocol.io/specification/2025-06-18/basic/lifecycle
• Rationale: The initialize result MUST carry a protocolVersion; without one no version is agreed.
• Requires transcript: False
• Reference: https://modelcontextprotocol.io/specification/2025-06-18/basic/lifecycle
• Remediation: Return a supported protocolVersion string from initialize.
• False-positive guidance: Not a false positive when the field is present but empty — fix the server.
• Fixture: mcp/unsafe/conformance/protocol-missing-protocol-version
• Scan modes: protocol, surface

protocol.missing-server-name

• Category: protocol
• Severity: error
• Spec version: 2025-06-18
• Spec reference: https://modelcontextprotocol.io/specification/2025-06-18/basic/lifecycle
• Rationale: serverInfo.name identifies the server to the host and MUST be present.
• Requires transcript: False
• Reference: https://modelcontextprotocol.io/specification/2025-06-18/basic/lifecycle
• Remediation: Set a non-empty serverInfo.name on initialize.
• False-positive guidance: Whitespace-only names are treated as missing.
• Fixture: mcp/unsafe/conformance/protocol-missing-server-name
• Scan modes: protocol, surface

protocol.missing-server-version

• Category: protocol
• Severity: warning
• Spec version: 2025-06-18
• Spec reference: https://modelcontextprotocol.io/specification/2025-06-18/basic/lifecycle
• Rationale: serverInfo.version lets a host pin and audit a server build; it should be declared.
• Requires transcript: False

protocol.protocol-version-downgraded

• Category: protocol
• Severity: info
• Spec version: 2025-06-18
• Spec reference: https://modelcontextprotocol.io/specification/2025-06-18/basic/lifecycle
• Rationale: The server negotiated an older revision than the client offered.
• Requires transcript: True

protocol.response-id-not-echoed

• Category: protocol
• Severity: error
• Spec version: 2025-06-18
• Spec reference: https://modelcontextprotocol.io/specification/2025-06-18/basic/transports
• Rationale: A response MUST echo the id of the request it answers, or it cannot be correlated.
• Requires transcript: True
• Reference: https://modelcontextprotocol.io/specification/2025-06-18/basic/transports
• Remediation: Echo the JSON-RPC request id on every successful or error response.
• False-positive guidance: Skipped when no transcript was captured; notifications (no id) are exempt.
• Fixture: mcp/unsafe/conformance/protocol-response-id-not-echoed
• Scan modes: protocol, requires_transcript

protocol.undeclared-capability-listed

• Category: protocol
• Severity: error
• Spec version: 2025-06-18
• Spec reference: https://modelcontextprotocol.io/specification/2025-06-18/basic/lifecycle
• Rationale: A server MUST NOT serve a capability it did not declare during initialize.
• Requires transcript: False
• Reference: https://modelcontextprotocol.io/specification/2025-06-18/basic/lifecycle
• Remediation: Declare the capability in initialize capabilities, or stop listing those items.
• False-positive guidance: Hosts that invent capability keys may need profile waivers — not silent skips.
• Fixture: mcp/unsafe/conformance/protocol-undeclared-capability-listed
• Scan modes: protocol, surface

protocol.unknown-capability-declared

• Category: protocol
• Severity: info
• Spec version: 2025-06-18
• Spec reference: https://modelcontextprotocol.io/specification/2025-06-18/basic/lifecycle
• Rationale: A capability key outside the spec's vocabulary belongs under 'experimental'.
• Requires transcript: False

protocol.unsupported-protocol-version

• Category: protocol
• Severity: error
• Spec version: 2025-06-18
• Spec reference: https://modelcontextprotocol.io/specification/2025-06-18/basic/lifecycle
• Rationale: The negotiated protocol version MUST be a revision this client speaks.
• Requires transcript: False
• Reference: https://modelcontextprotocol.io/specification/2025-06-18/basic/lifecycle
• Remediation: Negotiate a supported protocolVersion from the client's offered set.
• False-positive guidance: Bump Apiome's supported-version set only via an intentional release.
• Fixture: mcp/unsafe/conformance/protocol-unsupported-protocol-version
• Scan modes: protocol, surface

readiness.tool-description-too-brief

• Category: readiness
• Severity: warning
• Spec version: 2025-06-18
• Spec reference: https://modelcontextprotocol.io/specification/2025-06-18/server/tools
• Rationale: A description under 40 characters cannot distinguish a tool from its siblings, so an agent selects it by name alone.
• Requires transcript: False

readiness.tool-destructive-not-declared

• Category: readiness
• Severity: warning
• Spec version: 2025-06-18
• Spec reference: https://modelcontextprotocol.io/specification/2025-06-18/server/tools
• Rationale: A destructive operation that does not declare destructiveHint may be auto-approved by a host that would otherwise have demanded confirmation.
• Requires transcript: False

readiness.tool-missing-annotations

• Category: readiness
• Severity: info
• Spec version: 2025-06-18
• Spec reference: https://modelcontextprotocol.io/specification/2025-06-18/server/tools
• Rationale: Without behavioural annotations a host cannot reason about a tool's safety at all.
• Requires transcript: False

readiness.tool-missing-output-schema

• Category: readiness
• Severity: info
• Spec version: 2025-06-18
• Spec reference: https://modelcontextprotocol.io/specification/2025-06-18/server/tools
• Rationale: Without an outputSchema an agent cannot predict or validate a tool's result shape.
• Requires transcript: False

readiness.tool-missing-recovery-guidance

• Category: readiness
• Severity: info
• Spec version: 2025-06-18
• Spec reference: https://www.anthropic.com/engineering/writing-tools-for-agents
• Rationale: A description that never mentions the failure path leaves an agent with no recovery strategy when the call errors.
• Requires transcript: False

readiness.tool-name-unconventional

• Category: readiness
• Severity: info
• Spec version: 2025-06-18
• Spec reference: https://www.anthropic.com/engineering/writing-tools-for-agents
• Rationale: A name matching no common convention is hard for a model to reproduce exactly.
• Requires transcript: False

readiness.tool-naming-inconsistent

• Category: readiness
• Severity: info
• Spec version: 2025-06-18
• Spec reference: https://www.anthropic.com/engineering/writing-tools-for-agents
• Rationale: Mixed naming conventions across one server's tools make every name a guess.
• Requires transcript: False

readiness.tool-parameter-missing-description

• Category: readiness
• Severity: warning
• Spec version: 2025-06-18
• Spec reference: https://modelcontextprotocol.io/specification/2025-06-18/server/tools
• Rationale: An undocumented parameter forces an agent to infer its meaning from its name.
• Requires transcript: False

readiness.tool-parameter-unconstrained

• Category: readiness
• Severity: info
• Spec version: 2025-06-18
• Spec reference: https://modelcontextprotocol.io/specification/2025-06-18/server/tools
• Rationale: A free-text parameter with no enum, format, pattern, or bounds invites invalid arguments an agent cannot self-check.
• Requires transcript: False

readiness.tool-unbounded-list

• Category: readiness
• Severity: warning
• Spec version: 2025-06-18
• Spec reference: https://www.anthropic.com/engineering/writing-tools-for-agents
• Rationale: A collection-returning tool with no limit/cursor parameter can flood an agent's context with an unbounded result set.
• Requires transcript: False