Assistants

MCP connector

TRIVDA is a remote MCP server. Add one URL to your assistant, sign in, and it can search the licensed data your organisation may use, with the licence and source on every passage.

The connector URL

Server URLStreamable HTTP
https://13-63-0-55.sslip.io/connector/mcp

It is the host you sign in to TRIVDA on, followed by /connector/mcp (not /mcp, which is a dashboard page). The dashboard's Connectors page shows the exact URL with a copy button.

Two ways to authenticate, both charged to your organisation's wallet:

Claude

Claude on the web and Claude Desktop

  1. Open Settings → Connectors.
  2. Choose Add custom connector.
  3. Name it TRIVDA and paste the connector URL.
  4. Choose Add, then Connect, and approve TRIVDA in the window that opens.

A connector added on the web is also available in Claude Desktop with the same account, and the other way round. Custom connectors may need a paid Claude plan, and on Team and Enterprise plans an owner may have to add it for the organisation first.

Claude Code

Terminal
claude mcp add trivda --transport http https://13-63-0-55.sslip.io/connector/mcp

Then run /mcp in Claude Code, pick trivda and authenticate in the browser. To use an API key instead of OAuth:

TerminalAPI key
claude mcp add trivda --transport http https://13-63-0-55.sslip.io/connector/mcp \
  --header "Authorization: Bearer $TRIVDA_API_KEY"

ChatGPT

  1. Open Settings → Apps & Connectors.
  2. Turn on Developer mode under the advanced settings.
  3. Choose Create app.
  4. Name it TRIVDA and paste the connector URL.
  5. For authentication, choose API key and paste a TRIVDA API key (sk_live_...); ChatGPT sends it as Authorization: Bearer. OAuth also works if your workspace allows it.
  6. Choose Connect.

ChatGPT calls search to find passages and fetch to read them; reading a passage that a search already returned is free.

Cursor

  1. Open Cursor Settings → Tools & MCP and add a custom server, or edit ~/.cursor/mcp.json (all projects) or .cursor/mcp.json (one project).
  2. Add this and save:
mcp.json
{
  "mcpServers": {
    "trivda": {
      "type": "http",
      "url": "https://13-63-0-55.sslip.io/connector/mcp"
    }
  }
}
  1. Choose Login next to the trivda server and approve TRIVDA in the browser.

VS Code (GitHub Copilot)

  1. Run MCP: Add Server from the command palette and choose HTTP, or create .vscode/mcp.json in your workspace.
  2. Add this and save:
.vscode/mcp.json
{
  "servers": {
    "trivda": {
      "type": "http",
      "url": "https://13-63-0-55.sslip.io/connector/mcp"
    }
  }
}
  1. Start the server from the file or the MCP view, and approve TRIVDA when VS Code asks you to sign in. The tools appear in Copilot's agent mode.

Other clients

Any client that supports remote MCP servers over Streamable HTTP can connect: add a remote (HTTP) server with the connector URL. If it supports OAuth discovery it will open the TRIVDA sign-in by itself; if it asks for a key, use an API key. Menu names change between client releases; the URL is the only part that must be exact.

Tools

Search TRIVDA's licensed data. Charged like a REST search: per 1,000 tokens in the passages returned, nothing when nothing is found.

ArgumentTypeMeaning
query requiredstringWhat to search for, in plain language.
top_kinteger, default 8, max 50How many passages to return.
scopestringWhere to search: mine (your organisation's own data), marketplace (other organisations' published data), all (both) or listing:<id> (one listing from list_collections). Left out, your organisation's default applies; see Search scope.

It returns results, a list of passages; summary, the same results as readable text with the licence and source of each; scope, the scope the search ran in; cost, what the search cost (tokens, cost_eur, free_query_used, free_queries_remaining, balance_credits); and listings, the marketplace passages grouped by listing. Each result has:

id
<dataset_id>#<chunk>; pass it to fetch.
title, text
A short label and the passage.
url
The source document, for the citation.
licence
The SPDX licence id.
licence_duties, attribution
What the licence asks of you, and the attribution line to show with the passage.
provenance_id
Identifies this passage as served; keep it with anything you store or cite.
source
Who the data is from, as the partner chose to be shown.
origin, listing_id
own for your organisation's data, marketplace for a listing's; and the listing it belongs to.
score, dataset_id, published_date, readiness_tier
Relevance, where it came from, its date and its quality tier.
warning
Present when the owner labelled the passage fiction or synthetic. The assistant must not present it as fact.

When the wallet cannot cover a search, the tool returns a notice saying so instead of results, and nothing is charged.

fetch

Read one full passage by the id a search returned. Free for a passage your organisation was served in the last 7 days; otherwise it is charged like a one-passage search.

ArgumentTypeMeaning
id requiredstringAn id from search.

list_collections

List the listings a scope covers, so the assistant can choose where to look before it searches. Free.

ArgumentTypeMeaning
scopestringmine, marketplace, all or listing:<id>. Left out, your organisation's default applies.
querystringWords to narrow the list.
limitinteger, default 20, max 100How many listings to return.

It returns listings, scope and a readable summary. Each listing has a listing_id, title, origin (own or marketplace), source, licence, its size in passages, and scope: the value to pass to search to search only that listing. A listing never names the organisation behind it unless the partner chose to be shown.

get_balance

Show your organisation's balance and prices. Free; takes no arguments.

It returns credits (spendable), balance_eur, free_queries_remaining, currency, price_per_1k_tokens_eur (licensed data), own_data_per_1k_tokens_eur (your own data), default_scope and a readable summary. The assistant can use it to say what a search will cost or why one was refused.

Billing and limits

Troubleshooting