Connect Trove & deploy a toolkit

In a few minutes you’ll connect Trove to Claude and deploy your first toolkit, so Claude can search Wikipedia — and anything else in the catalog — right inside your conversation. (Every toolkit runs as a full MCP server on Trove’s cloud — no infrastructure on your side.)

Prerequisites

• A free Trove account — sign up.
• To deploy a toolkit: either Claude Code (the prompt path) or Node.js 20+ (the manual path).

1. Add Trove to Claude

Trove is one connection in Claude, added over its single endpoint (add it once from claude.ai on the web or the desktop app — the Claude mobile apps can’t add it, but once added, Trove works everywhere, phones included):

https://api.ontrove.sh/mcp

In Claude (web or desktop), open Settings → Connectors and click Add custom connector. Fill in:

• Name: Trove
• Remote MCP server URL: https://api.ontrove.sh/mcp

Leave Advanced settings collapsed — Trove registers your client automatically, so you don’t need an OAuth Client ID or Secret. Click Add.

Claude's Add custom connector dialog — Name and URL filled in, Advanced settings left collapsed.

Claude opens a browser window to sign in to Trove (create your account if you haven’t) and authorize the connection. That’s it — Claude now has Trove’s Core tools: trove_search, trove_recent, trove_save, and the rest.

To confirm, ask Claude:

What tools does Trove give you?

It lists the Core tools (trove_*). Your knowledge base starts empty — the next step gives Claude something to do.

Other MCP clients

Any MCP client that supports remote servers with OAuth works the same way — point it at https://api.ontrove.sh/mcp.

2. Deploy your first toolkit — Wikipedia

The official toolkits are open source in the toolkit catalog repo. Deploy the Wikipedia toolkit — it reaches a public API, so it needs no key. Do it two ways:

Ask Claude Code

If you use Claude Code, let it do the whole setup. Paste this prompt:

Install the Trove CLI (@ontrove/cli), run trove login, clone https://github.com/hollyburnanalytics/trove-integrations, then deploy its Wikipedia toolkit with trove mcp deploy --dir mcp/wikipedia. Finally run trove mcp ls to confirm it’s live.

Claude Code runs the commands in your terminal. When trove login opens your browser, sign in with the same Trove account. In under a minute, trove mcp ls shows wikipedia live.

Do it manually

Install the CLI and authenticate (the browser opens for the same account as Claude):

npm i -g @ontrove/cli   # Node 20+; installs the `trove` binary
trove login

Clone the catalog repo and deploy Wikipedia:

git clone https://github.com/hollyburnanalytics/trove-integrations.git
cd trove-integrations
trove mcp deploy --dir mcp/wikipedia

Confirm it’s running:

trove mcp ls

You’ll see wikipedia listed with its tools (search_articles, get_article).

Either way, Trove bundles the toolkit, uploads it, and activates it on its infrastructure — live in about a second.

3. Refresh the tool list and try it

Deployed toolkits join the same Trove connection — but if Claude is mid-session it won’t see them until you refresh. In Claude, open the Trove connector’s ⋮ menu and choose Refresh tools list (or just start a new chat). wikipedia__search_articles and wikipedia__get_article now appear.

Try:

Use Wikipedia to summarize the history of the printing press.

Claude calls the Wikipedia tools and answers from the live source. The toolkitId__toolName prefix keeps each toolkit’s tools distinct from the Core tools.

Skip the approval prompts

Under the connector’s Tool permissions, set Trove’s read-only tools to Always allow so Claude can use them without asking each time. (The other options are Needs approval, Blocked, and Custom.)

What the Core tools do

These ship with every Trove connection, alongside whatever toolkits you enable:

Prompt	Core tool
”Search my Trove for articles about distributed systems”	trove_search
”What’s new in my knowledge base?”	trove_recent
”Save this conversation to Trove”	trove_save
”Help me discover something interesting”	trove_discover
”Get the full text of that document”	trove_get_document
”List all my sources”	trove_list_sources

Deploy more

• Browse the catalog — the full catalog (arXiv, SEC EDGAR, FRED, the Met, and more). Deploy any of them the same way: trove mcp deploy --dir mcp/<slug>.
• Toolkits that need a key (Mapbox, FRED, X, eBay, OpenAlex, Resend) report a clear “not set” error until you set it: trove secret set <slug> <NAME> <value>.
• Build your own toolkit against @ontrove/mcp.

Troubleshooting

New tools not showing in Claude A deployed toolkit is live within ~a second. In Claude, open the Trove connector’s ⋮ menu → Refresh tools list (or start a new chat) to pull in the new tools.

trove login / authentication failed Re-run trove login to re-authorize. The CLI uses the same account as your Claude connection — no special access.

Session expired Sessions last about an hour; your client re-initializes automatically. No action needed.

Next steps

• Core Tools Reference — full documentation for the Core tools
• The Trove CLI — every command, and how it maps to the GraphQL API
• Authentication — how Trove authenticates clients via OAuth