A knowledge base and evidence management system with public search, editorial tools, member research capabilities, and MCP integration for evidence discovery and attribution
Base2026 tools are defined in a client-side JavaScript file (templates/base2026-tools-studio.js), not as MCP server tool registrations. The source shows UI panel definitions with labels, titles, and descriptions, but NO formal MCP tool schemas, input parameter type definitions, or output schemas. Tool names lack action verbs (find, extract, attribute, publish are weak). Descriptions exist but are UI-oriented, not LLM-optimized. Input parameters are mentioned in prose ('One narrow question', 'One or more public record IDs') but lack JSON Schema type definitions, constraints, or field-level descriptions. No error handling, no output schema documentation, no pagination guidance. This is a UI component, not an MCP server implementation.
Attribute every useful claim to its source.
Extract the passage without losing its context.
Find public evidence for one narrow question.
Build a brief. Choose the next move.
No JSON Schema input definitions visible. Tool parameters described in prose only ('One narrow question', 'One or more public record IDs') with no type, format, constraints, or field-level descriptions.
Tool names lack action verbs. 'find', 'extract', 'attribute', 'publish' are weak; should be 'search_evidence', 'extract_passage', 'attribute_claims', 'create_brief'. LLMs infer intent from verb-noun naming.
No output schema documented. Descriptions mention outputs ('Attributed records with a source path', 'Bounded excerpts, record metadata') but no structured field definitions, pagination, or chaining IDs for downstream tools.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | F | 28 | 2026-07-28+ | v2 |
Source code shows only a client-side UI component (JavaScript panel initialization), not an MCP server implementation. No evidence of MCP protocol transport, tool registration, or server-side logic.
No error handling, recovery guidance, or input validation documented. No mention of retryability, user-fixable errors, or actionable error messages.