Loading ToolBench
Connect Formik forms to AI assistants via WebMCP. Enables AI to fill fields, validate inputs, manage arrays, and navigate multi-step forms.
This adapter exposes 15 tools with templated names (e.g., get_{{formId}}_errors). While schemas are present and mostly well-formed, critical issues severely limit production readiness: (1) Tool names use template placeholders rather than concrete identifiers, making them non-functional as-is and preventing LLM tool selection. (2) Descriptions are minimal (10-40 chars) and lack context on WHEN to use each tool or what it returns. (3) No output schemas documented, LLMs cannot predict response structure. (4) No error handling guidance. (5) Tool definitions appear inferred from hook implementations rather than explicitly registered with full MCP ToolDefinition objects. The adapter is a React hook wrapper, not a standalone MCP server, it requires client-side integration and cannot function as a remote MCP server.
Add a new item to the {{fieldName}} array field of the {{formId}} form
Clear all items from the {{fieldName}} array field of the {{formId}} form
Get all current validation errors for the {{formId}} form
Get validation error for a specific field in the {{formId}} form
Get the current status of the {{formId}} form (dirty, submitting, valid, etc.)
Get the current step in the {{formId}} multi-step form
Get all items in the {{fieldName}} array field of the {{formId}} form
Jump to a specific step in the {{formId}} multi-step form
Tool names use unresolved template placeholders ({{formId}}, {{fieldName}}) instead of concrete identifiers. LLMs cannot select or invoke tools with placeholder names. Tools must be registered with actual form/field IDs at runtime.
Descriptions are extremely brief (10 - 30 characters) and lack context. They do not explain WHEN to use each tool, what it returns, or how it differs from similar tools. Example: 'Get all current validation errors for the {{formId}} form' is 50 chars but provides no guidance on return structure or use case.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | F | 28 | 2025-06-18+ | v2 |
Insert a new item at a specific index in the {{fieldName}} array field of the {{formId}} form
Move an item from one index to another in the {{fieldName}} array field of the {{formId}} form
Move to the next step in the {{formId}} multi-step form
Move to the previous step in the {{formId}} multi-step form
Remove an item at a specific index from the {{fieldName}} array field of the {{formId}} form
Swap two items in the {{fieldName}} array field of the {{formId}} form
Update an item at a specific index in the {{fieldName}} array field of the {{formId}} form
No output schemas documented. Tools return {content, structuredContent} but LLMs cannot predict the structure of structuredContent fields. This violates the pattern:response-shaper requirement and forces LLMs to guess at response structure.
No error handling or recovery guidance. Tools do not document what happens on failure (e.g., invalid field name, out-of-bounds step index, array full). LLMs receive no actionable error messages to guide retry or fallback logic.
This is a React hook adapter (FormikTools component + useFormikTools hook), not a standalone MCP server. Tool definitions are inferred from hook implementations rather than explicitly registered via MCP ToolDefinition. Cannot function as a remote MCP server without a transport layer (HTTP/SSE/STDIO).