Contract 1.7.0 · Local previewRead Markdown

Quickstarts

Local implementation verified. Hosted launch, live Claude and ChatGPT acceptance, and directory approval are pending. The production URL below is the planned endpoint. For testing, substitute the configured test endpoint supplied by its operator.

Claude connector

In Claude’s connector settings, add a custom remote connector using https://app.cadence-me.com/mcp. Complete Cadence’s Google sign-in and consent. Leave both write permissions off for a read-only trial. Ask for today’s Timeline; inspect the selected tools and returned dates.

Custom connector availability depends on your Claude account and organization policy. Cadence has not received directory approval. See Claude connector authentication.

Claude Code

Add the remote HTTP server:

claude mcp add --transport http cadence https://app.cadence-me.com/mcp

Open Claude Code and run /mcp to authenticate the Cadence connection. Complete the browser consent, then request the account timezone or Timeline. Do not paste an access token into a project configuration.

The command format follows Claude Code’s MCP documentation. Cadence-specific live acceptance remains pending.

Claude plugin

The plugin guide describes the local package, bundled skills and installation commands. The plugin uses the same endpoint and OAuth consent. Installing skills does not grant account access or bypass confirmation.

ChatGPT plugin and developer mode

In ChatGPT Plugins, choose the plus button, then Add custom MCP server. Enter Cadence and its endpoint, select OAuth, review the warning, and create the plugin. Authenticate with Cadence, review the tool list, then enable the plugin in a new conversation.

Older instructions may call this developer mode. The current documented flow starts in Plugins; account and workspace policies still control access. Follow OpenAI’s connection guide if the controls differ. This is setup guidance, not a claim that Cadence passed live ChatGPT acceptance. Packaged skills are covered under Plugins.

MCP Inspector

Use Inspector’s Streamable HTTP transport with the configured endpoint. Run discovery and OAuth, approve read access, list tools and call get_account_context with an empty object. Then try get_timeline with a date returned by the account context. Test writes only on disposable data with the corresponding permission enabled.

Inspector 2.11.0 completed local DCR, PKCE, consent and all 16 successful calls. Its error viewer can incorrectly validate an error against the success schema and report -32602. Inspect isError and the Cadence error code; official SDK clients preserve these errors. See troubleshooting.

Other clients

Use an MCP SDK with remote HTTP, OAuth discovery, PKCE S256, CIMD or DCR, and resource-bound tokens. Set resource to the exact MCP URL. Request cadence.read and only necessary write scopes. Request offline_access if refresh tokens are needed.

For the installed SDK v2 client, versionNegotiation mode auto selects the modern protocol; its default remains legacy. Discover schemas rather than copying them into prompts. A first call can use this SDK payload:

{"name":"get_account_context","arguments":{}}

This payload is not a complete HTTP request. The SDK supplies transport framing and headers.