Tools

The MCP server provides a set of tools that AI calls automatically when performing tasks.

Code Generation

generate_schema

Get template and rules for creating FormSchema.

Parameters:

Parameter	Type	Description
formType	string	Form type: "registration", "checkout", "settings", etc.

Returns:

• Type to component mapping (string → Input, boolean → Checkbox, etc.)
• Document masks for Russian documents (phone, INN, SNILS, passport)
• Schema structure rules
• Real project examples

---

generate_types

Get template and rules for TypeScript form types.

Parameters:

Parameter	Type	Description
formType	string	Form type for context

Returns:

• Interface naming rules
• Typing rules (undefined instead of null, no index signatures)
• Type examples

---

generate_validation

Get template and rules for ValidationSchema.

Parameters:

Parameter	Type	Description
formType	string	Form type for context

Returns:

• List of all built-in validators
• Usage syntax
• Import rules (@reformer/core/validators)
• Validation examples

---

generate_behavior

Get template and rules for BehaviorSchema.

Parameters:

Parameter	Type	Description
formType	string	Form type for context

Returns:

• List of all behaviors (computeFrom, enableWhen, etc.)
• Usage syntax
• Import rules (@reformer/core/behaviors)
• Behavior examples

---

Quality Checking

check_code

Analyze form code for errors and quality issues.

Parameters:

Parameter	Type	Required	Description
code	string	Yes	TypeScript code to analyze

Returns:

• Quality score (0-100 points)
• List of errors with explanations
• Warnings
• Fix recommendations

Checked rules:

• Correct validator and behavior imports
• Presence of value in every field
• Use of useMemo for createForm
• Call to markAsTouched on blur
• Type checking before submission

---

Development Assistance

explain_error

Explain a ReFormer error with fix examples.

Parameters:

Parameter	Type	Required	Description
error	string	Yes	Error text or problem description

Recognized issues:

• Field not updating on input
• Validation not triggering
• TypeScript type errors
• Form recreated on every render
• Cannot read property / undefined
• Array index out of bounds

---

get_pattern

Get usage pattern example.

Parameters:

Parameter	Type	Required	Description
pattern	string	Yes	Pattern name

Available patterns:

• conditional-fields — conditional field display
• computed-fields — computed fields
• cross-field-validation — cross-field validation
• async-validation — async validation
• array-fields — working with arrays
• multi-step-form — multi-step forms
• form-submission — form submission

---

get_recommended_structure

Get recommended project file structure.

Parameters:

Parameter	Type	Description
complexity	string	Complexity: "simple", "medium", "complex"

Returns:

• Folder structure
• File separation (types, schema, validation, behavior)
• Naming conventions

---

Feedback Collection

report_issue

Report an error and its solution for feedback collection.

Parameters:

Parameter	Type	Required	Description
error	string	Yes	Error message or problem description
solution	string	Yes	Solution or fix
tags	string[]	No	Tags for categorization and analytics
context	object	No	Additional context (see below)

Structure of context:

Field	Type	Description
examples	array	Code examples with descriptions
examples[].description	string	Example description
examples[].code	string	Example code
relatedFiles	string[]	Paths to related files
notes	string	Additional notes

Recommended tags:

Format	Example	Description
category:<type>	category:behavior, category:validation	Category: schema, validation, behavior, react, types, other
agent:<name>	agent:claude, agent:cursor	AI agent name
severity:<level>	severity:critical, severity:minor	Issue severity

Returns:

Confirmation of successful report submission.

How it works:

1. AI encounters an error while working with ReFormer
2. AI solves the problem
3. AI calls report_issue with error description and solution
4. Data is saved locally to ~/.reformer/issues.jsonl

Example usage:

AI found error: "Cycle detected in computeFrom"
AI identified cause: effect depends on target field
AI calls report_issue:
  - error: "Infinite loop in computeFrom when effect depends on target"
  - solution: "Use peek() instead of .value to read target without dependency"
  - tags: ["category:behavior", "agent:claude", "severity:critical"]
  - context:
      examples:
        - description: "Wrong - creates dependency"
          code: "const current = targetNode.value.value;"
        - description: "Correct - no dependency"
          code: "const current = targetNode.value.peek();"
      relatedFiles: ["packages/reformer/src/core/behavior/behaviors/compute-from.ts"]

Storage format (JSONL):

{"timestamp":"2025-01-15T10:30:00Z","error":"...","solution":"...","tags":["category:behavior"],"context":{"examples":[...],"notes":"..."}}

Each line is a separate JSON object. This allows easy appending and data analysis.

Why collect feedback?

Issue reports help:

• Identify common user problems
• Improve documentation
• Find library bugs
• Understand which patterns cause difficulties

All data is stored locally on your computer.