Skip to content

What is Bridge Town

Bridge Town is an AI-native financial planning and analysis (FP&A) platform. Instead of spreadsheets, you describe what you want in plain English and your AI agent — Claude, Codex, Gemini, or any other MCP-compatible client — writes Python financial models for you.

How it works

Section titled “How it works”
  1. You talk to your AI agent — describe the financial model you need (“Build a 12-month revenue forecast with three product lines”)
  2. Your agent uses Bridge Town’s MCP tools — creates a versioned project, writes Python model code, uploads your data
  3. Models run in a sandbox — isolated Docker containers with --network none, read-only /repo and /data, and writable tmpfs at /outputs and /tmp
  4. Results come back to your agent — stdout, output files, and dashboards are returned for analysis

Claude is the most heavily exercised path (Claude.ai uses OAuth, Claude Code and Claude Desktop use API tokens), but Bridge Town speaks plain MCP Streamable HTTP — any client that can attach an Authorization: Bearer … header works the same way. See Connecting any MCP Client for the generic flow.

Key concepts

Section titled “Key concepts”
  • Projects — Versioned workspaces managed by Bridge Town. Each project holds models, data sources, and dashboards.
  • Models — Python files stored at model/<name>.py in a project. Your agent writes them; you review and iterate.
  • Data sources — CSV, Excel, or Google Sheets data uploaded and stored as Parquet snapshots.
  • Branches — Project branches for scenario testing. Compare a “base case” against an “optimistic” scenario side-by-side.
  • Sandbox execution — Models run in Docker containers with --network none, a read-only root filesystem, read-only /repo + /data, and writable tmpfs for /outputs + /tmp.

Architecture

Section titled “Architecture”
┌─────────────┐   MCP/HTTP   ┌──────────────┐     Gitea API    ┌─────────┐
│  Any MCP    │ ◄──────────► │  MCP Server  │ ◄──────────────► │  Gitea  │
│  agent      │              │  (FastAPI)   │                  │  (Git)  │
└─────────────┘              └──────────────┘                  └─────────┘
   Claude.ai                        │
   Claude Code                      ▼
   Claude Desktop          ┌──────────────┐
   Codex / OpenCode        │   Postgres   │
   Custom agents           │  (RLS/RBS)   │
                           └──────────────┘
  • MCP Server — FastAPI application exposing tools via the Model Context Protocol (Streamable HTTP transport)
  • Gitea — Internal project storage (implementation detail). Each tenant gets an org; each project maps to a Git repository underneath.
  • PostgreSQL — Tenant data with Row-Level Security. Every query is scoped to the authenticated tenant.
  • Redis — Rate limiting and Celery task broker
  • Docker sandbox — --network none containers for model execution

Who is it for?

Section titled “Who is it for?”
  • Finance teams who want AI-assisted model building without giving up version control
  • Developers building financial tooling on top of a secure, audited platform
  • AI agents (Claude, Codex, Gemini, OpenCode, custom MCP clients, …) that need structured tools for financial modeling tasks