# Bridge Town MCP Connector Build financial models as code with cloud execution, Native Sheets, Google Sheets, version control, and collaboration. This public repository contains documentation, registry metadata, a minimal stdio-to-HTTP shim, and versioned model-authoring helpers for the hosted Bridge Town MCP connector. The commercial server implementation is operated by Bridge Town and is not published here. - Website: - App: - Documentation: - MCP endpoint: `https://api.bridgetown.builders/mcp` - Transport: MCP Streamable HTTP - Authentication: OAuth for hosted connector flows; bearer tokens for manual client configuration Bridge Town does not run a server-side LLM or proxy prompts. The user's chosen AI client and model provider run the conversation; Bridge Town receives typed MCP calls and executes deterministic platform operations. ## Connect For Claude.ai, add `https://api.bridgetown.builders/mcp` as a custom connector and approve the Bridge Town OAuth prompt. For clients that support Streamable HTTP plus headers: ```json { "mcpServers": { "bridge-town": { "transport": { "type": "http", "url": "https://api.bridgetown.builders/mcp", "headers": { "Authorization": "Bearer btk_YOUR_TOKEN" } } } } } ``` Generate and revoke API tokens at . See [Getting Started](docs/getting-started.md) for setup, current task flows, and troubleshooting. ## Canonical MCP surface The production endpoint exposes one canonical tool catalog. Use `search_tools` to find the correct current tool for a goal and `get_tool` to retrieve its full schema. The catalog does not change based on the client or user agent. The canonical registry manifest for this public listing is [server.json](server.json). It points to the hosted remote MCP server; it is not a copy of the proprietary server implementation. ## Versioned model-authoring helpers Small helper modules that are intended to run inside a model repository are published as immutable, MIT-licensed releases: - [`native_sheet_reader.py` v1.1.0](model-authoring-helpers/v1.1.0/native_sheet_reader.py) preserves Native Sheet missing/blank/formula states and emits bounded, value-free read identities. - [`output_lineage.py` v1.1.0](model-authoring-helpers/v1.1.0/output_lineage.py) emits validated `output_lineage.json` dependency graphs. The [v1.1.0 release directory](model-authoring-helpers/v1.1.0/) includes runnable examples, a changelog, ownership, license, manifest, and `SHA256SUMS`. Install a helper byte-for-byte at the manifest's `target_path` and verify its digest before committing it to a model repository. ## Local stdio shim Most clients should connect directly to the Streamable HTTP endpoint. A local MCP host that requires stdio may use the minimal forwarding shim: ```bash BRIDGE_TOWN_API_TOKEN=btk_YOUR_TOKEN npx github:Bridge-Town/financial-modeling-mcp ``` The shim does not read or write local files or spawn child processes. It only forwards MCP JSON-RPC messages over HTTPS to the configured Bridge Town MCP endpoint. ## Security Financial model execution runs in an isolated container without outbound network access. Google access uses `drive.file`, limiting Bridge Town to files selected through the Bridge Town flow or created by Bridge Town. Report vulnerabilities privately through [SECURITY.md](SECURITY.md). Do not open a public issue or include customer data in a report. ## Governance - [Support](SUPPORT.md) - [Contributing and contribution boundary](CONTRIBUTING.md) - [Public and private source boundary](SOURCE_SCOPE.md) - [Ownership](OWNERS.md) - [MIT license](LICENSE) The governance files are generated from the closed, versioned [`directory-resources.v1.json`](directory-resources.v1.json) metadata contract.