Budget Planner MCP
Enables budget management, scenario creation, and value updates through natural language, integrating with the Budget Planner MVP.
README
Budget Planner MVP
A TypeScript prototype for a monthly budget planning application. The backend is intentionally small and uses Node.js built-in HTTP only, with all domain and calculation logic kept outside the server layer so it can be replaced later.
Project Structure
apps/backend/ Node HTTP JSON API
apps/frontend/ React + Vite + ReactGrid UI
packages/domain/ Budget model, seed data, formula evaluation, recalculation
The domain package owns the core concepts: Budget, Scenario, Series, Value, and FormulaBinding. React components never evaluate formulas.
Setup
Install dependencies:
pnpm install
Run backend and frontend together:
pnpm dev
Environment variables are loaded from a root .env file. Start from one of the examples:
cp .env.dev.example .env
To use local passwordless user-name login during development:
ALLOW_PASSWORDLESS_AUTH=true VITE_ALLOW_PASSWORDLESS_AUTH=true pnpm dev
Default URLs:
- Frontend:
http://localhost:5173 - Backend:
http://localhost:4000
Set APP_BASE_PATH and VITE_APP_BASE_PATH to serve the app under a subpath such as /abc. With VITE_APP_BASE_PATH=/abc, the frontend root is http://localhost:5173/abc/, and scenario pages use URLs such as /abc/budgets/budget-mvp/scenarios/scenario-base.
From the root page, users can create a new budget or create a new scenario by copying an existing scenario in the same budget. New budgets start with a Base Case scenario containing only Income Subtotal, Expense Subtotal, and Total.
Scenario pages subscribe to Server-Sent Events at ${APP_BASE_PATH}/api/budgets/:budgetId/scenarios/:scenarioId/events, so edits from another browser or API client update open views automatically.
Local development can use a passwordless user-name auth mode. It is disabled by default and must be enabled with ALLOW_PASSWORDLESS_AUTH=true for the backend and VITE_ALLOW_PASSWORDLESS_AUTH=true for the frontend. Do not enable this mode for public production traffic. Budget owners can share budgets with other users from the root page.
The backend also accepts Google OAuth access tokens with Authorization: Bearer <token>. Tokens are validated against Google's userinfo endpoint and mapped to the verified email address.
Run individual apps:
pnpm --filter backend dev
pnpm --filter frontend dev
Typecheck all packages:
pnpm typecheck
Build all packages:
pnpm build
API
API routes live under APP_BASE_PATH. With APP_BASE_PATH=/abc, use /abc/api/....
GET /api/statereturns the seeded budget, scenario, periods, series, values, and formula bindings.PUT /api/valuesupdates one editable monthly value and recalculates affected downstream series for the same period.PUT /api/series/:seriesId/formulaupdates one series formula and bindings, checks circular dependencies, and recalculates affected values.POST /api/recalculaterecalculates the full scenario for maintenance/debugging.
Remote MCP
The backend exposes a prototype remote MCP endpoint at POST ${APP_BASE_PATH}/mcp. It supports JSON-RPC MCP methods initialize, tools/list, and tools/call.
Authentication options:
- Optional passwordless mode: send
X-Dev-User: aliceonly whenALLOW_PASSWORDLESS_AUTH=true. - Remote Google login: send
Authorization: Bearer <Google OAuth access token>.
OAuth protected resource metadata is available at:
GET /.well-known/oauth-protected-resource
GET /.well-known/oauth-protected-resource/mcp
When APP_BASE_PATH=/abc, these are served as /abc/.well-known/oauth-protected-resource and /abc/.well-known/oauth-protected-resource/mcp.
Available MCP tools:
list_budgetsget_scenario_stateupdate_valuefor one value or bulk updates withupdates: [{ seriesId, period, value }]create_seriesfor one series or bulk creation withseries: [{ key, name, type, formula, unit, initialValues }]update_seriesdelete_seriescreate_budgetcreate_scenarioshare_budget
When using MCP to add income or expense series, prefer a unit-based structure instead of a single manual amount. Add parameter series for drivers such as users, quantity, headcount, unit price, or unit cost, then add the income/expense series with a formula such as users * unitPrice or headcount * unitCost whenever that structure can be reasonably inferred. For bulk creation, put parameter series before formula series that reference them, and use initialValues to seed monthly driver values in the same MCP call.
MCP updates reuse the same domain logic and publish the same SSE scenario updates as the browser UI.
Production Deployment
Use Google sign-in for production and leave passwordless auth off:
cp .env.prod.example .env
pnpm deploy:prod
The backend can serve HTTPS directly when certificate files are already available:
HTTPS=true TLS_CERT_FILE=/etc/letsencrypt/live/example.com/fullchain.pem TLS_KEY_FILE=/etc/letsencrypt/live/example.com/privkey.pem PORT=443 pnpm --filter backend start
For automatic Let's Encrypt issuance and rotation, use Caddy in front of the Node backend:
BUDGET_DOMAIN=example.com BUDGET_BASE_PATH=/abc ACME_EMAIL=admin@example.com BACKEND_UPSTREAM=localhost:4000 FRONTEND_DIST=/srv/budget-frontend caddy run --config deploy/Caddyfile
Set BUDGET_BASE_PATH to the same value as APP_BASE_PATH; use an empty value for domain-root deployment or /abc for subpath deployment. Caddy serves the built frontend over HTTPS, renews certificates automatically, and proxies ${BUDGET_BASE_PATH}/api/*, ${BUDGET_BASE_PATH}/mcp, and OAuth metadata requests to the backend.
For pm2-based deployments, pnpm deploy:prod builds all workspaces, replaces the contents of FRONTEND_DIST, and restarts the backend with pm2 restart ${PM2_APP_NAME}. Configure both values in the root .env file:
FRONTEND_DIST=/var/www/abc
PM2_APP_NAME=budget-backend
Set DEPLOY_SKIP_PM2_RESTART=true if you only want to rebuild and sync the frontend output without restarting pm2.
Seed Scenario
The in-memory store starts with MVP Budget, currency JPY, from 2026-01 to 2026-06, and a Base Case scenario.
Seeded series include Revenue, Users, Unit Price, Engineer Cost, Headcount, Unit Cost, Cloud Cost, Income Subtotal, Expense Subtotal, and Total. Revenue, Engineer Cost, subtotals, and Total are calculated automatically from formula bindings.
Formulas support arithmetic, variables, parentheses, and aggregate expressions such as Sum(Type=Income) and Sum(Type=Expense).
The grid display order is derived from domain meaning and formula bindings: income rows and their subtotal, expense rows and their subtotal, Total, other calculated rows, then unreferenced parameters. Parameters referenced by a formula are shown under the referencing series in formula variable order.
Notes
No database, ORM, authentication, or backend web framework is used. All data resets when the backend process restarts.
Recommended Servers
playwright-mcp
A Model Context Protocol server that enables LLMs to interact with web pages through structured accessibility snapshots without requiring vision models or screenshots.
Magic Component Platform (MCP)
An AI-powered tool that generates modern UI components from natural language descriptions, integrating with popular IDEs to streamline UI development workflow.
Audiense Insights MCP Server
Enables interaction with Audiense Insights accounts via the Model Context Protocol, facilitating the extraction and analysis of marketing insights and audience data including demographics, behavior, and influencer engagement.
VeyraX MCP
Single MCP tool to connect all your favorite tools: Gmail, Calendar and 40 more.
graphlit-mcp-server
The Model Context Protocol (MCP) Server enables integration between MCP clients and the Graphlit service. Ingest anything from Slack to Gmail to podcast feeds, in addition to web crawling, into a Graphlit project - and then retrieve relevant contents from the MCP client.
Kagi MCP Server
An MCP server that integrates Kagi search capabilities with Claude AI, enabling Claude to perform real-time web searches when answering questions that require up-to-date information.
E2B
Using MCP to run code via e2b.
Neon Database
MCP server for interacting with Neon Management API and databases
Exa Search
A Model Context Protocol (MCP) server lets AI assistants like Claude use the Exa AI Search API for web searches. This setup allows AI models to get real-time web information in a safe and controlled way.
Qdrant Server
This repository is an example of how to create a MCP server for Qdrant, a vector search engine.