<!-- CURSOR_SUMMARY -->
> [!NOTE]
> **Medium Risk**
> Touches multiple toolkits’ runtime entrypoints and context/error/auth
plumbing, so breakage risk is mainly around invocation/packaging and
tool execution wiring rather than business logic.
>
> **Overview**
> Migrates the BrightData, ClickHouse, LinkedIn, Math, MongoDB,
Postgres, and Zendesk OSS toolkits from `arcade-tdk` to
`arcade-mcp-server` APIs by updating tool decorators, `Context` types,
auth classes, and exception imports.
>
> Adds per-toolkit `__main__.py` files that construct an `MCPApp`,
register module tools, and run via configurable transport/host/port;
corresponding `pyproject.toml` updates bump versions, drop
`arcade-tdk`/`arcade-serve` deps, and add `project.scripts` console
entrypoints.
>
> Updates tests and eval suites to use `arcade_mcp_server.Context`
(mocked) and switches eval `ToolCatalog` imports to `arcade_core`.
>
> <sup>Written by [Cursor
Bugbot](https://cursor.com/dashboard?tab=bugbot) for commit
9b3e31acb4b35e1d72efd47e2d279c5b19e3ecb0. This will update automatically
on new commits. Configure
[here](https://cursor.com/dashboard?tab=bugbot).</sup>
<!-- /CURSOR_SUMMARY -->
As of today, Arcade's toolkits are moving to closed-source.
Toolkit source code remains available upon request for our paying
customers. The history of this repository contains the toolkits as of
their most recent publication under their previous license terms.
Adds an example of a good "general case" SQL tool:
* enforces read-only mode
* hints to the LLM to discover the tables and schemas for the tables it
needs before any query
* uses RetryableToolErrors to hint to the LLM about what to do next
Docs: https://github.com/ArcadeAI/docs/pull/345
For testing, `TEST_POSTGRES_DATABASE_CONNECTION_STRING` has been set in
the repo (from Neon). details in 1 password.
<img width="1178" height="1091"
alt="464977013-49aff5e5-e301-4ca0-83b5-3ea742db2283"
src="https://github.com/user-attachments/assets/9344c27b-015d-4b91-907e-84f2e4193e16"
/>
