pilates_mcp
This server is a local-first MCP server for managing Pilates center promotional posters and schedule workspaces, with optional PNG/PDF rendering and Railway deployment support.
Create, update, import, search, reuse, and archive editable HTML/CSS promotional posters
Render posters to PNG/PDF using headless Chromium at Instagram/A4 sizes
Manage center profile, poster defaults, themes, moods, styles, and design feedback
Recommend poster styles based on local archive history and saved feedback
Store and retrieve schedule settings, weekly templates, and center closures
Import, inspect, generate, edit, validate, save, and export group/private XLSX schedules
Find available schedule slots and detect schedule conflicts
Create schedule posters from public-safe schedule data, excluding private names by default
Operate via stdio MCP or HTTP JSON-RPC, with OAuth-ready Railway endpoints and a lightweight /schedule web workspace
Click on "Deploy Server".
Wait a few minutes for the server to deploy. Once ready, it will show a "Started" state.
In the chat, type
@followed by the MCP server name and your instructions, e.g., "@pilates_mcp create a poster for our new year special promotion"
That's it! The server will respond to your query, and you can continue using it as needed.
Here is a step-by-step guide with screenshots.
pilates_mcp
pilates_mcp is a local-first MCP server for Pilates center promotional posters. It creates editable HTML/CSS poster sources, stores them in a local archive, searches and reuses past posters, records design feedback, and can render PNG/PDF artifacts with headless Chromium.
Phase 1 deliberately does not store member names, phone numbers, schedules, health data, payment data, or other member personal information.
Project Structure
This repository was empty at implementation time, so the server is implemented as a TypeScript + Node.js MCP server.
src/
server.ts MCP stdio JSON-RPC transport and tool dispatch
service.ts poster workflow orchestration
repository.ts local poster archive search/read/write
renderer.ts Playwright PNG/PDF renderer
templates.ts HTML/CSS template layer
feedback.ts design feedback storage
profile.ts center profile storage
themes.ts design token presets
types.ts domain types
data/
posters/YYYY/<poster-id>/
poster.html
poster.css
metadata.json
content.json
preview.png optional
poster.pdf optional
revisions/ update backups
feedback/
global.json
table.json
standard.json
poster-<poster-id>.json
config/
center-profile.json
schedules/
settings.json
center-closures.json
group/
template.json
group_schedule.xlsx
revisions/
private/
template.json
private_schedule.xlsx
revisions/Related MCP server: Jusratio Case File
Install
npm install
npx playwright install chromiumDevelopment
npm run build
npm test
npm run devPILATES_MCP_DATA_DIR can point the archive to another local data directory. If unset, the server uses ./data.
Railway Deployment
Railway deployment is configured with railway.json and Dockerfile.
The production container uses the official Playwright image so Chromium runtime dependencies are available for PNG/PDF rendering. Railway starts npm run start:railway, which exposes:
GET /health: deployment healthcheckGET /tools: MCP tool summary for deployment inspectionGET /schedule: lightweight Excel-style Schedule WorkspaceGET /mcp: MCP HTTP endpoint infoPOST /mcp: JSON-RPC MCP endpoint forinitialize,tools/list, andtools/call
The stdio MCP server remains available through npm run start:stdio for MCP clients that connect by command. Railway uses npm start and POST /mcp for HTTP JSON-RPC MCP calls.
For persistent local-first archive storage on Railway, mount a Railway volume at /data or set PILATES_MCP_DATA_DIR to another persistent path.
Set these Railway variables for ChatGPT remote MCP OAuth:
PILATES_MCP_OAUTH_PASSWORD: owner password shown during/authorizeapprovalPILATES_MCP_PUBLIC_URL: canonical public origin, for examplehttps://your-service.up.railway.appPILATES_MCP_AUTH_DISABLED: optional local-only escape hatch; set totrueonly for local development
PILATES_MCP_PUBLIC_URL must use the same origin as the MCP URL configured in ChatGPT. For example, if ChatGPT uses https://pilates-mcp.com/mcp, set PILATES_MCP_PUBLIC_URL=https://pilates-mcp.com. Do not mix it with https://mcp.pilates-mcp.com unless that subdomain is the actual MCP URL and all well-known OAuth endpoints are served from that same subdomain.
OAuth endpoints:
/.well-known/oauth-protected-resource/.well-known/oauth-protected-resource/mcp/.well-known/oauth-authorization-server/.well-known/openid-configuration/register/authorize/token
If authorization finishes and the browser shows a 502 for a URL like http://127.0.0.1:<port>, that URL is the MCP client's local OAuth callback, not the Railway service. Check that the ChatGPT/Codex client connection flow is still active and that local callback URLs are not blocked by browser, firewall, proxy, or VPN settings.
MCP Client Connection
Use the built server as a stdio MCP command:
{
"mcpServers": {
"pilates_mcp": {
"command": "node",
"args": ["G:/pilates_mcp/dist/src/server.js"],
"env": {
"PILATES_MCP_DATA_DIR": "G:/pilates_mcp/data"
}
}
}
}For HTTP-capable MCP clients, point the client at:
https://<railway-domain>/mcpExample JSON-RPC call:
curl -X POST https://<railway-domain>/mcp \
-H "content-type: application/json" \
-d '{"jsonrpc":"2.0","id":1,"method":"tools/list","params":{}}'MCP Output Schemas
Every MCP tool publishes an outputSchema through tools/list. This tells ChatGPT and other MCP clients what structured JSON shape to expect before they call a tool.
Tool calls also return structuredContent alongside the human-readable JSON text response. The text response is kept for compatibility, while structuredContent is the preferred machine-readable result for agents. For example, create_poster can return valid: false with missing fields and questions, search_posters returns an items array, and schedule mutation tools return success, conflicts, and saveState.
This matters for this server because many tools are workflow tools, not one-shot generators. The schema helps the agent understand when it must ask for missing poster settings, when explicit schedule save is still required, and how to safely inspect conflicts or render artifact paths.
Tools
Poster tools:
get_center_profileupdate_center_profilesearch_postersget_postercreate_posterupdate_posterreuse_poster_templateimport_posterrender_posteradd_design_feedbackget_design_feedbackrecommend_poster_style
Schedule tools:
get_schedule_settingsupdate_schedule_settingsinspect_schedule_fileimport_schedule_filecreate_schedule_templateget_schedulegenerate_weekly_scheduleset_schedule_slotclear_schedule_slotfind_available_slotsvalidate_scheduleadd_center_closureremove_center_closureget_center_closuressave_scheduleexport_schedulecreate_schedule_poster
Poster Creation Flow
create_poster reads the center profile, applies center defaults, validates the PosterBrief, gathers relevant feedback, resolves the layout and theme, renders static HTML/CSS, writes metadata/content/source files, and optionally renders PNG/PDF.
The required order is:
user request
-> center profile defaults
-> PosterBrief
-> validation
-> ask only missing questions
-> past posters and feedback
-> HTML/CSS
-> optional render
-> archive saveRequired poster brief decisions are validated before generation:
eventNameprogramSeparationtablePreferencetableStylewhen a table is required or providedvatPolicywhen price content exists
The server returns valid: false, missing, and Korean questions instead of guessing missing event names, VAT policy, or program separation.
programSeparation and tableStyle are separate decisions. For example, group/private products can use programSeparation: "table" with tableStyle: "split" to render separate GROUP/PRIVATE table blocks. tablePreference represents user intent; layout is the final renderer choice.
Center defaults can reduce repeated questions:
{
"posterDefaults": {
"programSeparation": "table",
"tablePreference": "required",
"tableStyle": "split",
"vatPolicy": "excluded"
}
}Priority is: current request and PosterBrief, then center profile defaults, then a validation response asking the user for only missing values. VAT is never inferred.
Examples
8월 회원권 가격표 포스터 생성
{
"title": "8월 회원권 이벤트",
"purpose": "event",
"brief": {
"eventName": "8월 회원권 이벤트",
"programCategories": ["private", "duet"],
"programSeparation": "table",
"tablePreference": "required",
"tableStyle": "comparison",
"vatPolicy": "excluded"
},
"content": {
"subtitle": "이번 달 특별 혜택",
"badge": "8월 EVENT",
"table": {
"columns": ["상품", "정상가", "이벤트가"],
"rows": [
["1:1 개인레슨 10회", "800,000원", "650,000원"],
["듀엣레슨 10회", "500,000원", "390,000원"]
],
"highlightColumn": 2
},
"footer": "선착순 마감"
},
"style": "premium",
"render": { "png": true, "pdf": true, "size": "instagram-portrait" }
}작년 9월 포스터 틀 재사용
{
"sourcePosterId": "poster_20250901_ab12cd34",
"title": "2026년 9월 신규회원 이벤트",
"brief": {
"eventName": "9월 신규회원 이벤트",
"vatPolicy": "excluded"
},
"content": {
"subtitle": "올해 가격으로 업데이트",
"table": {
"columns": ["상품", "정상가", "이벤트가"],
"rows": [["1:1 개인레슨 10회", "820,000원", "670,000원"]],
"highlightColumn": 2
}
}
}최근 1년 기록 기반 스타일 추천
{
"purpose": "event",
"month": 9,
"recentMonths": 12
}Feedback
Feedback is stored in JSON files, not hard-coded. create_poster retrieves active global and layout-scoped feedback and includes it in the result as appliedFeedback.
Example:
{
"scope": "table",
"text": "이벤트가는 정상가보다 확실히 강조."
}Rendering
render_poster uses Playwright Chromium with JavaScript disabled. Supported sizes:
instagram-portrait: 1080 x 1350instagram-square: 1080 x 1080story: 1080 x 1920a4-portrait: A4 portrait
HTML/CSS are the source of truth. PNG/PDF are optional rendered artifacts.
Schedule Workspace
Phase 2 adds a local-first Schedule Workspace for Pilates group/private timetables. It is not a CRM. Private schedules may contain displayName only for grid display and operational placement.
Core model:
ScheduleSettings: week display, opening/closing time, default slot minutes, holiday option, locale, country codeScheduleTemplate: explicit-save weekly group/private templateScheduleSlot: neutral domain slot separate from Excel cellsScheduleTemplateMapping: XLSX worksheet/header/time/day-column mappingCenterClosure: center-specific closed dates, separate from national holidays
set_schedule_slot changes an in-memory dirty workspace. It is not persisted until save_schedule is called. export_schedule writes XLSX and backs up an existing XLSX before replacing it.
Existing XLSX files are handled by inspect_schedule_file and import_schedule_file. The importer captures worksheet name, used range, day columns, time column, column widths, row heights, and merges where available. Formulas are not executed.
Holiday handling is provider-based. Phase 2 includes a local provider with basic KR holiday fixtures and can be replaced later by an external provider without changing the domain model. Center closures have priority over public holidays, but occupied slots are not silently deleted; warnings are returned.
Railway serves a lightweight Excel-style workspace at /schedule with sticky headers, direct cell editing, delete/backspace, and visible save state. MCP remains the source of actual schedule operations.
Schedule Examples
Scenario A, create a new group timetable when no file exists:
{
"scheduleType": "group",
"settings": {
"weekDisplay": "mon-sat",
"openingTime": "09:00",
"closingTime": "21:00",
"defaultSlotMinutes": 60,
"autoHolidayMarking": true,
"locale": "ko-KR",
"countryCode": "KR"
}
}Scenario B, import an existing Excel file and generate next week:
{ "filePath": "fixtures/group_schedule.xlsx", "scheduleType": "group" }{
"scheduleType": "group",
"weekStart": "2026-10-05",
"applyHolidays": true,
"applyCenterClosures": true
}Scenario C, private schedule operation:
{
"scheduleType": "private",
"dayOfWeek": "monday",
"startTime": "19:00",
"displayName": "김OO"
}Scenario D, group schedule poster:
{
"scheduleType": "group",
"weekStart": "2026-10-05",
"title": "9월 그룹레슨 시간표",
"render": { "png": true, "pdf": true, "size": "instagram-portrait" }
}Security And Privacy
The archive uses generated poster IDs and validates poster IDs before file access. Archive-relative imports are checked to block ../ traversal and root escape. Generated poster HTML is static, user text is HTML-escaped, and remote scripts are not part of the template. Rendering disables JavaScript.
Phase 1 excludes member management, reservations, payments, SMS, Instagram posting, StudioMate integration, external design scraping, image generation AI, cloud DB requirements, and SaaS authentication.
Phase 2 permits only displayName as person-related schedule display data. It does not create fields for phone, email, address, birth date, gender, health condition, medical history, payment, membership, consultation, attendance profile, or marketing consent. Names are not used for profiling, analytics, poster feedback, marketing, or recommendations.
Phase 2 still excludes StudioMate integration, StudioMate scraping, Chrome extension automation, member CRM, contact management, payment management, attendance analysis, member preference profiling, automatic booking confirmation, SMS, KakaoTalk, Instagram posting, payroll, settlement, and complex ERP features.
Extension Points
The service/repository split keeps room for future semantic search or another index. External design references can be added later behind a provider interface such as:
interface DesignReferenceProvider {
search(query: string): Promise<DesignReference[]>;
}Phase 1 does not implement scraping or copying external templates.
Available Tools
29 toolsadd_center_closureA
Add a center-specific closure date. If existing schedule entries are present, returns warnings and requires confirmation semantics without deleting data.
Inputs: date (required), label.
add_design_feedbackB
Store reusable design feedback locally. Scope can be global, table, standard, or a specific poster.
Inputs: text (required), scope (required), active, posterId.
clear_schedule_slotA
Clear a schedule slot in the dirty workspace without saving until save_schedule is called.
Inputs: startTime (required), scheduleType (required), date, dayOfWeek.
create_posterA
Create and archive a new HTML/CSS poster only after required operating and price-display rules are known. Validates PosterBrief first and returns missing questions instead of guessing eventName, VAT policy, table preference, table style, or program separation.
Inputs: title (required), content (required), purpose (required), mood, +11 more.
create_schedule_posterA
Create a Phase 1 poster from public-safe schedule data. Private displayName values are excluded by default.
Inputs: title (required), scheduleType (required), render, weekStart.
create_schedule_templateA
Create a new unsaved weekly group/private schedule template after required settings are known. Does not save until save_schedule is called.
Inputs: scheduleType (required), settings.
export_scheduleA
Export the current schedule to XLSX, preserving mapping metadata where possible and backing up an existing XLSX before writing.
Inputs: scheduleType (required), filename.
find_available_slotsA
Find available schedule slots by schedule type, date/day, time range, and duration.
Inputs: scheduleType (required), date, toTime, duration, +2 more.
generate_weekly_scheduleB
Generate an unsaved weekly schedule from the weekly template plus holidays, center closures, and overrides.
Inputs: weekStart (required), scheduleType (required), applyHolidays, applyCenterClosures.
get_center_closuresB
Return locally stored center-specific closure dates.
Inputs: no named parameters.
get_center_profileA
Return the local Pilates center profile and poster defaults. Use before poster generation.
Inputs: no named parameters.
get_design_feedbackC
Return active local design feedback by scope, poster, or layout so generation can reference it.
Inputs: scope, layout, posterId, activeOnly.
get_posterA
Return one archived poster's metadata, structured content, theme, HTML, CSS, and local archive paths.
Inputs: posterId (required).
get_scheduleB
Return the current dirty draft or saved schedule template.
Inputs: scheduleType (required).
get_schedule_settingsA
Return local non-CRM schedule settings such as week display, opening hours, slot minutes, locale, and holiday marking.
Inputs: no named parameters.
import_posterB
Import existing HTML/CSS into the local archive with metadata. Image-only poster reconstruction is intentionally out of Phase 1 scope.
Inputs: metadata (required), css, html, theme, +3 more.
import_schedule_fileA
Import an existing group/private XLSX schedule into a dirty Schedule Workspace draft while preserving mapping metadata where possible.
Inputs: filePath (required), scheduleType.
inspect_schedule_fileA
Inspect an existing XLSX schedule file inside the workspace and infer worksheet, used range, day columns, time column, merges, widths, and heights without executing formulas.
Inputs: filePath (required).
recommend_poster_styleB
Recommend poster layout/style/table structure from local archive history and saved feedback only. Does not search the internet.
Inputs: month, purpose, recentMonths, programCategories.
remove_center_closureB
Remove a center-specific closure date.
Inputs: date (required).
render_posterB
Render an archived HTML/CSS poster to PNG and/or PDF with headless Chromium. JavaScript is disabled during rendering.
Inputs: posterId (required), pdf, png, size.
reuse_poster_templateB
Copy an existing poster's layout/style/theme into a new poster and replace only the content. Never overwrites the source poster.
Inputs: title (required), content (required), sourcePosterId (required), mood, +11 more.
save_scheduleA
Explicitly save the dirty schedule template and create a revision backup when replacing an existing saved template.
Inputs: scheduleType (required).
search_postersC
Search local archived poster metadata by date, layout, purpose, style, feedback-related tags, free text, VAT policy, and table structure.
Inputs: tag, mood, year, limit, +11 more.
set_schedule_slotB
Set one group/private schedule slot in the dirty workspace. Conflicts are returned unless overwrite is explicitly true. Only displayName is allowed for private names; no CRM fields.
Inputs: startTime (required), scheduleType (required), date, note, +6 more.
update_center_profileA
Update non-member Pilates center profile fields and default poster expression rules such as VAT and table style.
Inputs: name, brand, logoPath, footerText, +3 more.
update_posterA
Update an existing poster source HTML/CSS through structured content/theme fields. Creates a simple revision backup first.
Inputs: posterId (required), mood, notes, style, +4 more.
update_schedule_settingsC
Update center schedule defaults. These are operational settings and must not include member CRM data.
Inputs: locale, closingTime, countryCode, openingTime, +3 more.
validate_scheduleA
Validate duplicate slots, operating hours, holiday conflicts, time ranges, and Excel mapping warnings.
Inputs: scheduleType (required).
Tool Schema Changelog
Recent tool additions, removals, and schema changes observed during successful MCP inspections.
Loading tool schema changelog…
TDQS
Scored across 29 tools
Most tools map to distinct resource-action pairs, and the descriptions generally make boundaries clear. The closest potential confusions are create_schedule_poster vs. create_poster and inspect_schedule_file vs. import_schedule_file, but each has enough distinction to guide selection.
All tool names consistently follow a snake_case verb_noun pattern, such as get_schedule, update_center_profile, render_poster, and validate_schedule. There are no mixed naming conventions or vague one-word tool names.
29 tools is above the comfortable range for a single MCP server and crosses the 'too many' threshold. While the tools are organized into settings, schedule, and poster workflows, the large surface increases selection burden and makes the server feel heavy.
The tool surface covers center profile settings, schedule import/template/validation/save/export, and poster creation/archive/render/feedback/search comprehensively. Minor lifecycle gaps exist, such as no explicit delete operation for a poster or an entire saved schedule template, but these are workaroundable.
Maintenance
Related MCP Connectors
HTML and CSS to PDF MCP server with page headers, footers and page numbers. No headless browser.
API-first CRM for LLMs - contacts, companies, deals and activities over a native MCP server.
Self-hosted AI workspace: semantic RAG document search, long-term memory, and AI chat via MCP.
WordPress MCP server: publish posts, AI images, SEO and full site management, self-hosted
Related MCP Servers
- AlicenseAqualityDmaintenanceMCP server that turns articles, transcripts, and markdown into LinkedIn carousel PDFs, Instagram PNGs, and Threads PNGs. Content in, slides out. No web UI, no cloud service.53MIT
- AlicenseBqualityAmaintenanceLocal-first MCP server that transforms civil case PDFs into queryable cases with structural provenance, evidence IDs, and page-level verification for drafting and factual review.22Apache 2.0
- AlicenseNot gradedqualityBmaintenanceLocal-first personal productivity MCP server bundling email, calendar, messaging, search, and image generation tools with BYOK and no cloud dependency.94 npmMIT
- AlicenseBqualityCmaintenanceA local-first open-source MCP server that acts as a personal secretary, memory manager, and automation hub—it syncs Calendar, Gmail, Canvas, GitHub, and Health into a typed temporal memory graph, and enables approval-gated write actions like creating events, drafting/sending mail, and filing issues, all with telegram/slack/CLI interfaces and a SQLite-backed audit trail.33Apache 2.0