AI assistants (MCP)

Shrimp Tide runs a Model Context Protocol (MCP) server, so AI assistants such as Claude, ChatGPT, Cursor and Claude Code can work in your workspace: log time, start and stop the timer, look up projects and tasks, and run reports. Assistants use the same operations as the REST API, with the same validation, permissions and audit log.

Before you start

MCP access is part of full API access on the Team and Enterprise plans. The server lives at /api/mcp on the same origin as the app:

https://app.example.com/api/mcp

Owners find the URL and copy-ready setup snippets in the app on the API keys page.

Connecting with your account

Remote connectors such as claude.ai and ChatGPT sign in with your Shrimp Tide account. Add a custom connector with the server URL, then:

1. Sign in to Shrimp Tide.
2. Choose the workspace the assistant may use, if you belong to more than one.
3. Choose the access level and select Allow.

The assistant then acts as you, with your role in that workspace — it can never do more than you can in the app. Claude Code connects the same way; run this command and follow the sign-in link:

claude mcp add --transport http shrimp-tide https://app.example.com/api/mcp

Connecting with an API key

Clients that let you set request headers — Claude Code, Cursor, scripts — can use an organization API key instead. As on the REST API, the key acts like an owner of the workspace, and its access level decides which tools the assistant sees.

claude mcp add --transport http shrimp-tide https://app.example.com/api/mcp \
  --header "Authorization: Bearer $KEY"

Cursor and most other clients read a JSON configuration such as .cursor/mcp.json:

{
  "mcpServers": {
    "shrimp-tide": {
      "url": "https://app.example.com/api/mcp",
      "headers": { "Authorization": "Bearer shrimp_…" }
    }
  }
}

Access levels

When you connect with your account, you choose the level on the consent screen; with an API key, it comes from the key. Your role in the workspace still limits what every tool may do.

Tools

Every REST endpoint is also a tool, named after its router and procedure, for example timeEntries_create or projects_list. Read-only tools and tools that delete data are marked as such, so assistants can ask before they act. Invalid input or missing permissions come back to the assistant as readable tool errors.

Connected apps

Under Account → Connected apps you see every assistant you connected and can revoke its access. Owners see and revoke the connections of all members on the API keys page. Revoking takes effect immediately, even for tokens the assistant already holds.

Each connection may make 1,000 requests per hour; after that the server answers 429 with a Retry-After header. New and revoked connections appear in the audit log.

Next steps

Every operation behind the tools, with its input schema, is listed in the interactive reference:

• API reference
• API keys