Author Style "-esque" MCP Server

A curated catalog of 11 author writing styles decomposed into 8 orthogonal dimensions, with dual-output paths for text generation and image generation. Each style is an independent "stompbox" that colors prompts with structural writing patterns — not copied text.

Architecture

Layer 1: Taxonomy (0 tokens)     → author_style_taxonomy.py
Layer 2: Deterministic (0 tokens) → author_style_operations.py
Layer 3: Creative synthesis       → Consumer LLM responsibility
Server:  FastMCP interface        → author_style_mcp.py

Layer 1 holds the pure data — 11 author coordinates in 8D style-space, dimension specifications, and output vocabulary mappings. Layer 2 performs all deterministic operations: distance computation, weighted interpolation (blending), vocabulary extraction, and prompt generation. No LLM calls at any point. Layer 3 is left to the consuming application — a single Claude or other LLM call that uses the structured directives as creative input.

The Catalog

ID	Style	Origin	Signature
hemingway	Hemingway-esque	English (American)	Iceberg theory, paratactic flatness, submerged tension
de_sade	Marquis de Sade-esque	French	Baroque nesting, exhaustive enumeration, philosophical excess
le_guin	Ursula K. Le Guin-esque	English (American)	Balanced cadence, anthropological worldbuilding, warm precision
didion	Joan Didion-esque	English (American)	Clinical observation, specific sensory detail, retrospective present
lovecraft	Lovecraft-esque	English (American)	Accumulative horror, archaic register, cosmic scale
borges	Borges-esque	Spanish (Argentine)	Labyrinthine logic, infinite recursion, philosophical miniatures
murakami	Murakami-esque	Japanese	Flat affect, mundane surrealism, domestic loneliness
marquez	Márquez-esque	Spanish (Colombian)	Magical realism, multigenerational fate, tropical profusion
kafka	Kafka-esque	German (Czech)	Bureaucratic absurdism, plain surface over impossible premise
shonagon	Sei Shōnagon-esque	Japanese (Classical)	List-form observation, radical sensory specificity, aesthetic judgment
lispector	Clarice Lispector-esque	Brazilian Portuguese	Interior stream, philosophical viscerality, self-examining language

The 8 Dimensions

All values normalized [0.0, 1.0].

Dimension	Low End	High End	Text Output	Image Output
Syntactic Density	Paratactic / flat	Hypotactic / nested	Sentence length, clause depth	Compositional layering depth
Sensory Concreteness	Abstract / conceptual	Concrete / sensory	Noun register, verb type	Material rendering specificity
Ornamental Register	Stripped / minimal	Lush / baroque	Adjective density, figurative language	Surface detail complexity
Tension Visibility	Submerged / iceberg	Externalized / explicit	Show vs. tell ratio	Lighting drama, contrast ratio
Tension Temporality	Ruptural / episodic	Accumulative / inevitable	Pacing, foreshadowing density	Temporal framing, motion state
Reality Stability	Unstable / paradoxical	Stable / verifiable	Epistemic mode, hedging language	Spatial logic, physics accuracy
Interiority	Exterior / behavioral	Interior / consciousness	POV mode, thought access	Framing distance, depth of field
Temporal Mode	Eternal present / episodic	Cyclical / exhaustive	Tense, temporal scope	Motion blur, temporal compositing

Tools

Layer 1 — Taxonomy Lookup (0 tokens)

• get_author_styles() — List all 11 authors with coordinates
• get_author_style_profile(author_id) — Complete profile: coordinates, signature moves, text/image vocabulary
• get_style_dimensions() — All 8 dimensions with low/mid/high output mappings
• get_parameter_names() — Ordered parameter list for dynamics integration

Layer 2 — Deterministic Operations (0 tokens)

• compute_author_distance(author_id_1, author_id_2) — Euclidean distance with per-dimension breakdown
• blend_author_styles(blend_spec_json) — Weighted interpolation of multiple styles
• generate_text_style_prompt(author_id | blend_spec_json | custom_coordinates_json) — Text-generation directives
• generate_image_style_prompt(author_id | blend_spec_json | custom_coordinates_json, style_modifier) — Image-generation visual vocabulary
• find_style_extremes() — Maximum-contrast pair across catalog
• find_nearest_style(author_id) — Closest neighbor in style-space

Usage Examples

Single style — text prompt

generate_text_style_prompt(author_id="kafka")

Returns structured directives: sentence architecture, vocabulary constraints, forbidden words, paragraph rhythm, and a composited master prompt string.

Single style — image prompt

generate_image_style_prompt(author_id="marquez", style_modifier="photorealistic")

Returns keywords, color palette, compositional rules, and per-dimension visual directives suitable for ComfyUI / Stable Diffusion / DALL-E.

Blended style

blend_author_styles('{"hemingway": 0.6, "borges": 0.3, "shonagon": 0.1}')

Interpolates in 8D style-space. Returns blended coordinates, nearest catalog author (Didion, distance 0.406), merged signature moves, and extracted text/image vocabularies for the blend point.

Distance analysis

compute_author_distance("hemingway", "lovecraft")

Returns Euclidean distance (1.863), per-dimension breakdown, and identifies the maximum-contrast axis (ornamental_register, gap 0.80).

Dynamics Integration

Coordinates are directly compatible with aesthetics-dynamics-core tools:

# Get Hemingway coordinates
coords = get_coordinates("hemingway")
# → {"syntactic_density": 0.10, "sensory_concreteness": 0.90, ...}

# Use with integrate_trajectory for smooth style morphing
integrate_trajectory(
    start_state=get_coordinates("hemingway"),
    target_state=get_coordinates("lovecraft"),
    parameter_names=PARAMETER_NAMES,
    num_steps=30
)

Also composable with catastrophe-morph and surface-design-aesthetics servers — stack an author style brick with a catastrophe type or surface treatment for cross-domain aesthetic composition.

Composition Example: The Stompbox Chain

User prompt
  → Hemingway-esque (text structure: terse, concrete, submerged)
  → Catastrophe Morph: cusp (visual geometry: sharp vertices, crystalline)
  → Surface Design: matte_ceramic (material: chalky, light-absorbing)
  → Final synthesis (single LLM call)

Each brick is independent, deterministic, and zero-cost. Creative synthesis happens once at the end.

Deployment

FastMCP Cloud entry point:

author_style_mcp.py:mcp

Local execution:

python author_style_mcp.py

File Structure

author_style_taxonomy.py    # Layer 1: Types, dimensions, author catalog
author_style_operations.py  # Layer 2: Distance, blend, prompt generation
author_style_mcp.py         # FastMCP server with tool decorators
pyproject.toml              # Package configuration
README.md                   # This file

Style-Space Topology Notes

From validation testing:

• Maximum contrast pair: Hemingway ↔ Lovecraft (1.863) — wider than Hemingway ↔ de Sade (1.792) due to Lovecraft's combined extremes on ornament, reality stability, and interiority
• Surprising neighbors: Kafka ↔ Le Guin (0.636) — structurally closer than their reputations suggest, sharing balanced syntactic density and mid-range tension handling
• Emergent blends: 60/30/10 Hemingway/Borges/Shōnagon lands in Didion's neighborhood — concreteness + interiority + sensory precision ≈ cool clinical observation
• Unique coordinate: Sei Shōnagon occupies a region no other author approaches — low density + extreme concreteness + episodic temporality + high interiority is a combination absent from the Western literary tradition