Skip to main content
Works in Claude, Codex & Cursor

Plan in plain English. Ship in Python.

Bridge Town turns financial models into a version-controlled codebase your agent can read, run, and review. Every change is controlled, tracked, and tested.

Run the product walkthrough Why financial models belong in code Bring your own AI. Bridge Town brings security, memory, and collaboration.
bridgetown install connector ready
Add Bridge Town to Claude in one click.

Open Claude with Bridge Town pre-configured as a connector. OAuth — no tokens to copy. Chat seamlessly across desktop, web, and mobile in the same session.

  • Sign in with Google or Microsoft. No credit card. Sandbox workspace ready in ~6 seconds.
  • Authorise Bridge Town. Bridge Town's OAuth consent screen appears — sign in and approve access to your workspace.
  • Ask Claude anything. "Build me a three statement model for our new Business Unit." It replies, with a link back to the Bridge Town report.
Need a different tool? Full install docs → Free for the first workspace
/ what you get

A model your agent can read, run, and review.

Bridge Town isn't a wrapper around a spreadsheet. It's a financial modeling platform where every assumption is code, every scenario is a branch, and every actual is a commit.

/ proof from the top

Bridge Town runs in Claude, Codex, Cursor, and any MCP client.

One install, four agents, same answers. Your team picks the surface they're already in. The model is the source of truth.

Claude Codex Cursor OpenCode + any MCP
What's our Q3 EBITDA variance vs plan?
Bridge Town · fetch_variance(period: "Q3-2026", metric: "ebitda")
Q3 EBITDA came in +$1.4M (+8.2%) over plan, driven by NA enterprise.
/ version control

Every model change is a pull request.

Branch, propose, review, merge. The merge gate is the EBITDA delta — not "looks good to me."

/ audit-grade

Provenance baked in, not bolted on.

Every number traces to the commit, the scenario, the input source, the analyst. Auditors stop asking; you stop screenshotting.

/ how it works

Three commands. Three minutes.

From an empty workspace to a model your agent can query — fast enough to do during a stand-up.

/ step 01

Install in your agent.

Click Connect to Claude, or paste the install command into Codex, Cursor, or OpenCode. OAuth in one window.

$ connect bridgetown
connector ready
/ step 02

Describe a model.

"Build a 3-statement model for an SaaS company doing $40M ARR, 110% NRR, 65% gross margin." Your agent generates the Python through Bridge Town MCP tools; Bridge Town runs it sandboxed and shows the result.

build_model(arr: $40M, nrr: 1.10, gm: 0.65)
✓ generated 12 drivers · 36 periods
/ step 03

Review, branch, ship.

Open a PR for next quarter's headcount plan. The reviewer sees the EBITDA delta, the driver-level attribution, and the run log — before they hit merge.

PR #142 · headcount-q3-conservative
EBITDA Δ −$0.4M · Cash Δ +$2.1M
Get started

Install once. Ask anywhere.

Connect Bridge Town to the agent your team already lives in. Free for your first workspace; production teams scale to seats.

Install
claude mcp add bridgetown
Ask
"Q3 variance by department?"
See variance
EBITDA Δ +$1.4M · Cash Δ −$0.3M
Security

Your numbers stay your numbers.

Bridge Town is built for finance teams who can't afford a leak. Encryption at every layer, scoped access for every agent, and an audit log that reads like git history — because it is git history.

/ 01 · encryption

Encrypted in transit and at rest.

TLS enforced on every wire. AES-256 on every disk. Per-workspace data keys in AWS KMS, rotated quarterly.

  • TLS (HSTS enforced)
  • AES-256-GCM at rest
  • Per-workspace KMS envelope
  • Quarterly key rotation
/ 02 · access

Workspace-scoped tokens, not blanket keys.

Every agent connection uses a workspace-scoped token tied to a single account. Tokens carry the issuing user's permissions and are revocable from the dashboard at any time.

  • OAuth 2.1 with PKCE
  • SSO via Google & Microsoft
  • Per-workspace token isolation
  • Revocable from the dashboard
/ 03 · audit

Every change is a traceable commit.

Models, runs, and scenario edits land as commits in an append-only log. Every number traces to a commit, a scenario, an input source, and an analyst.

  • Append-only commit log
  • Full model and run history
  • CloudWatch audit delivery
  • Exportable log records