# Roadmap

This roadmap outlines the planned direction of the Utsuho project.

Utsuho aims to remain a **lightweight, deterministic, pure-Python
library for Japanese text normalization**.

The project focuses on:

- deterministic character-level transformations
- minimal dependencies
- predictable behavior suitable for preprocessing and search
  normalization

Utsuho intentionally avoids dictionary-based processing and heavy
linguistic frameworks.

## Near Term

### MCP Interface

Build on the newly added Model Context Protocol (MCP) interface so that Utsuho can be used more effectively from AI systems and developer tools.

Delivered in 2.2.0:

- MCP server implementation
- four conversion tools:
  - `half_to_full`
  - `full_to_half`
  - `hiragana_to_katakana`
  - `katakana_to_hiragana`
- optional dependency (`utsuho[mcp]`)
- stdio transport support
- width-conversion options exposed for MCP clients

Near-term improvements:

- richer MCP usage examples and client integration guides
- clearer compatibility guidance for MCP hosts and agent environments
- possible support for additional MCP capabilities such as resources or
  prompts, where they fit the project's deterministic scope

This allows AI systems and agents to normalize Japanese text using
Utsuho while keeping the interface explicit and predictable.

### Kana → Romaji Conversion

Add deterministic kana-to-romaji conversion.

Goals:

- deterministic mapping
- no dictionary or morphological analysis
- suitable for indexing and search preprocessing

Expected features:

- Hepburn style output
- handling of:
  - small kana
  - long vowels
  - sokuon (っ)
  - yōon (きゃ, しゃ, etc.)

## Mid Term

### Iteration Mark Expansion

Add support for expanding Japanese iteration marks.

Examples:

    時々 → 時時
    いろゝ → いろろ
    サヽキ → ササキ

Supported characters:

    ゝ ゞ ヽ ヾ

This feature remains deterministic and does not require dictionaries.

### Performance Improvements

Continue improving performance while keeping the implementation pure
Python.

Possible improvements include:

- optimized string building
- reduced allocations
- faster lookup paths

Benchmarks will be maintained to detect performance regressions.

## Long Term

### API Stabilization

Review and refine the public API as the project grows.

Possible improvements include:

- improved configuration design
- clearer separation between normalization types
- better extensibility for future converters

Breaking API changes, if necessary, would be introduced in a future
major release.

## Non-Goals

To keep the project focused and maintainable, Utsuho intentionally
avoids:

- dictionary-based processing
- morphological analysis
- heavy linguistic frameworks
- native extensions

The project will remain a **pure-Python deterministic normalization
library**.