Skip to content

Quick Start

Get from zero to a working financial model in four steps.

Sign up at app.bridgetown.builders. You’ll land on the Dashboard after login.

Use the sidebar and go to Models (/models) to view your models.

A model is a versioned workspace that holds your files, data, and dashboards.

Bridge Town speaks the Model Context Protocol (MCP) — bring whichever agent you prefer. Claude.ai is the fastest path because it can connect over OAuth with no API token.

Your selected provider processes the AI conversation under your agreement. Bridge Town does not host or invoke the language model; it receives the MCP tool calls your client sends.

Claude.ai (recommended, no token):

  1. Click Customize+Add custom connector
  2. Paste https://api.bridgetown.builders/mcp
  3. Authorise the connection when prompted

See Connecting Claude.ai for full details.

For everything else, generate an API token from Connect AI at app.bridgetown.builders/connect and follow the relevant guide:

The supported-client roster below is generated from the Connect AI catalog. Open Connect AI for the exact command, configuration path, and verification step.

ClientAuthenticationSetup
Claude.ai (claude-ai)oauthOAuth connection
ChatGPT (chatgpt)oauthOAuth connection
Claude Code (claude-code)tokencommand
Claude Desktop (claude-desktop)tokenconfig
Codex (codex)tokencommand and config
Cursor (cursor)tokenconfig
GitHub Copilot (copilot)tokencommand and config
Gemini CLI (gemini-cli)tokencommand
Windsurf (windsurf)tokenconfig
Amp (amp)tokencommand
Mistral Vibe (mistral-vibe)oauthOAuth connection
OpenCode (opencode)tokencommand and config
Kimi Code (kimi-code)tokencommand and config
Z Code (z-code)tokengui
Qwen Code (qwen-code)tokencommand and config
Cline (cline)tokenconfig
Goose (goose)tokenconfig
Aider (aider)tokenunsupported
Droid (droid)tokencommand

The MCP tool surface is identical across clients. The Anthropic Messages API MCP connector is tools-only; full MCP clients can also surface Bridge Town resources and prompts.

Open your agent and ask it to build a model:

“Create a 12-month revenue forecast in my forecasts model. Use three product lines: SaaS, Services, and Marketplace. Start from January 2026.”

Your agent will:

  1. Call create_model if the model does not exist
  2. Call commit_files to write the Python model and supporting files
  3. Call run_model with mode="sync" to execute it in the sandbox
  4. Return the results (stdout and output files)

You can then iterate:

“Add a 10% annual growth rate to the SaaS line”

Your agent reads the model, calls patch_file or commit_files to apply the change, then calls run_model again.