Capy MCP Server
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., "@Capy MCP Serverdraft an option grant for our new engineer with 10,000 shares"
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.

Why Capy
We were Pulley customers too. We wanted a small, understandable product we could keep running, with code we could inspect and data we could export. So we built Capy.
This repository contains the web app, backend functions, equity calculations, import/export code, and deployment configuration. The hosted service at capyinc.com runs this application. Self-hosted installations do not require a Capy subscription.
Related MCP server: Gogi MCP Server
What you can do
Workflow | Included |
Bring your cap table | Preview a detailed Pulley Excel export before signup; review reconciliation and warnings before saving; import contacts separately. |
Manage equity | Track stakeholders, shares, options, SAFEs, share classes, and equity plans. Record corrections, option exercises, and cancellations with activity history. |
Understand ownership | Review fully diluted ownership, plan reserves, and supported monthly vesting schedules. Search, sort, and filter the cap table. |
Model fundraising | Compare priced-round dilution with pre-money and post-money SAFEs, caps, and discounts; save scenarios. |
Share with stakeholders | Give a stakeholder access to their own holdings through an invitation that can be claimed once and revoked. |
Keep records | Store documents and save approval, offer, and communication drafts. |
Take your data | Download the cap table as Excel and individual tables as CSV. |
Use it from AI apps | Connect Claude, ChatGPT, Cursor, or another AI app over MCP to read the cap table and draft changes for an admin to apply. |
Model your next round

Track your SAFEs

Screenshots use fictional companies and synthetic records.
Current scope
Capy is a young project with a working hosted product. We publish the implementation, checks, and limitations so you can evaluate it yourself.
The current release does not send documents for signature, move money, file taxes, provide 409A appraisals, or automate legal-document generation. Open Cap Table Format export is on the roadmap; exports today are Excel and CSV. Email verification/password recovery delivery and administrator-team invitations are not yet implemented.
Read supported workflows and boundaries and the Pulley import guide before migrating a company.
Use Capy from your AI app
Add https://capyinc.com/mcp to Claude, ChatGPT, Cursor, VS Code or another MCP client, sign in, and choose which companies the app can see. The app can read your cap table and, if you allow it, draft grants, exercises, cancellations, stakeholder updates and board consents. Nothing changes until an admin applies the draft in Capy. See connecting AI apps.
Run locally
Requirements: Bun 1.3.14, Node.js 24, and your own Convex project. Stripe and Cloudflare credentials are not needed for local development.
git clone https://github.com/zain/capy.git
cd capy
bun install --frozen-lockfile
cp apps/web/.env.example apps/web/.env.local
bun run dev:setupCreate or select your own Convex project. Once the CLI has written packages/backend/.env.local, use a second terminal from the repository root to configure its development environment:
cd packages/backend
bunx convex env set SITE_URL http://localhost:3001
bunx convex env set BETTER_AUTH_SECRET "$(openssl rand -hex 32)"
bunx convex env set CAPY_SELF_HOSTED true
cd ../..The setup command waits for a successful push; if it reports missing environment variables, leave it running while setting them. Copy the deployment's .convex.cloud and .convex.site URLs into apps/web/.env.local, then:
bun run devOpen localhost:3001, create an account, and import a workbook or create an empty company. See the development guide for the full environment reference and troubleshooting.
Checks
bun run check # formatting, lint, production build, and TypeScript
bun run test # import/export, calculations, billing, and webhook tests
bun audit # known dependency vulnerabilitiesGitHub Actions runs these checks on pushes and pull requests, with a frozen lockfile and no production credentials. See CONTRIBUTING.md for the review and testing expectations.
Architecture
Part | Technology |
Web app | React, TanStack Start, TypeScript |
Database and server functions | Convex |
Authentication | Better Auth with the Convex adapter |
Calculations and validation | Decimal.js and Zod |
Excel import/export | ExcelJS |
Tooling | Bun and Vite+ |
Hosted web deployment | Cloudflare Workers |
apps/web/ Routes, product UI, authentication, imports, exports, and the MCP server
packages/backend/ Convex schema, functions, permissions, and file storage
packages/equity/ Shared equity calculations, validation, and modeling
packages/ui/ Shared UI components
packages/infra/ Cloudflare deployment command
scripts/ Production build helper
docs/ Setup, architecture, product scope, and migration guidesRead the architecture guide for data flow and access boundaries. The self-hosting guide covers deployment on your own Cloudflare and Convex accounts, updates, and backups. Running the Convex server itself on your infrastructure is not yet a tested deployment path for this repository.
Maintainers and support
Built by Zain and Sina. Zain maintains this repository; hello@capyinc.com reaches the team.
Bugs and feature requests: open an issue.
Setup and usage questions: Discussions.
Private migration or account help: email us.
Security vulnerabilities: use private vulnerability reporting, as described in SECURITY.md.
Please use synthetic examples in public issues. See SUPPORT.md for what to include in a report.
License and acknowledgments
Capy is licensed under GNU AGPL v3.0 (AGPL-3.0-only). Contributions use the same license. Dependency and asset licenses remain with their respective authors; see THIRD_PARTY_NOTICES.md.
Built with Better T Stack, TanStack, Convex, Better Auth, and shadcn/ui. Capy is an independent project and is not affiliated with Pulley or Carta.
This server cannot be deployed
Maintenance
Related MCP Connectors
- ZapierOAuthcom.zapier.mcp
Zapier MCP connects AI tools like Claude, ChatGPT, and Cursor to over 8,000 apps and 30,000+ actions, enabling AI to perform real-world tasks such as sending messages, searching data, scheduling events, and updating records. It acts as a translator between AI tools and apps, handling authentication, rate limits, and retries automatically, transforming AI from a conversational tool into a functional extension of your business stack.
Let AI agents query data and act across all your business apps via MCP.
- Cavuno MCPOAuthcom.cavuno
Connect Claude, Cursor, Codex, and other MCP clients to manage your Cavuno job board.
- QuallaaOAuthcom.quallaa
Talk to your public-facing AI from any MCP client — Claude, ChatGPT, Cursor, Cline, Windsurf.
Related MCP Servers
- AlicenseNot gradedqualityBmaintenanceConnect MCP-compatible AI assistants like Claude, Codex, and ChatGPT to your InterviewFlowAI workspace so they can search candidates, review interview scores and details, and update supported candidate fields through permission-aware tools.2MIT

Gogi MCP Serverofficial
AlicenseNot gradedqualityDmaintenanceEnables AI assistants in MCP clients to query real-time market data, inspect portfolio and Web3 wallet balances across brokers, place protected equity/ETF, crypto, forex, options, DeFi swap, and prediction market orders, and run backtests.MIT- AlicenseBqualityAmaintenanceEnables AI hosts like Claude, ChatGPT, Cursor, Codex, and Copilot to connect to DealDesk for working with cards, quotes, CPQ portfolio, directory, and export.6037 npmMIT
- AlicenseNot gradedqualityAmaintenanceConnects any MCP-capable AI assistant—ChatGPT, Claude, Gemini, Cursor and others—to an AI-native CRM via OAuth, so users can manage contacts, deals, pipelines, reports and add-ons by chat, with every action scoped to the connecting user's role. The server itself never calls an AI model or holds API keys, leaving all intelligence to the assistant you bring.19 npmAGPL 3.0