Finalize application to Anthropic

Release date: 26 August 2026

This release improves MCP compatibility, tool discoverability and metadata quality across supported AI platforms. This release prepares RED for review in Anthropic’s Claude Connector Directory while preserving all existing accounting workflows and safety controls.

Improved MCP tool descriptions

All 159 production tools now publish concise, factual descriptions of their purpose, required inputs, outputs, effects and limitations.

Previous descriptions occasionally included conversational guidance or directions about when another tool should be used. These have been replaced with neutral API-contract wording, improving compatibility with MCP directory requirements and making RED’s capabilities easier for AI clients to interpret.

Stronger description-policy validation

RED now applies a fail-closed policy whenever tools are registered. A tool cannot be published if its description contains:

• Instructions that attempt to control AI-model behaviour

• Directions to invoke or avoid another tool

• Conversational or answer-formatting instructions

• Hidden, encoded or non-printing text

• Other metadata that does not belong in a factual tool description

Descriptions are validated without silently deleting or rewriting their contents, making future metadata issues easier to detect during development.

Updated help-tool metadata

The descriptions for RED’s Big Red Cloud help and resource-discovery tools have been simplified and made more factual. RED continues to provide:

• Big Red Cloud help articles

• Step-by-step product guidance

• Freshdesk resources

• BRC Edu and webinar content

• Screenshot-supported instructions

• Public support links

These help workflows remain read-only and can be used without connecting a company.

Improved connector- directory compatibility

RED’s production metadata has been verified through a live MCP tools/list exchange.

The production catalogue now confirms:

• 159 production tools

• 159 unique tool names

• 159 compliant, non-empty descriptions

• Matching top-level and annotated titles

• Complete read-only, destructive and open-world safety annotations

• No hidden, encoded or behaviour related instructions in tool descriptions

This work supports RED’s application for inclusion in Anthropic’s Claude Connector Directory. The application is currently under review; directory availability remains subject to Anthropic’s approval.

Existing safety controls remain unchanged

This release does not change RED’s accounting logic or posting behaviour.

RED continues to:

• Display previews before supported accounting writes

• Require explicit confirmation before posting

• Require additional confirmation for destructive and external actions

• Validate route tokens and company context

• Preserve counterparty and concurrency safeguards

• Scope accounting actions to connected companies

• Prevent RED from initiating real bank transfers or moving financial assets
  
  

Customer impact

No customer action is required.

Existing RED connections and supported accounting workflows continue to operate as before. 

Quality assurance

RED 1.6.3 was validated through:

• Full application test coverage: 1,074 passed and 1 intentionally skipped

• Focused MCP registration and description-policy tests

• TypeScript type checking

• Production and staging MCP metadata audits

• Claude help, read-only and confirmation-boundary smoke tests

• Live verification of all 159 production tool descriptions

No accounting records were changed as part of the production metadata audit.