OrionBI is an OAuth-protected Model Context Protocol (MCP) server that gives your AI client governed, audited access to your team's databases, materialized data sets, and GitHub — with group and org tool governance and a full audit trail. This guide covers connecting it to Claude, ChatGPT, Cursor, or any MCP client, and what each tool does.

Before you start

You need:

An OrionBI account with product access — an active trial or paid seat, or an exempt account. Sign up at orionbi.ai.
At least one resource configured in the admin — a database connection, a data set, or a connected GitHub account — so the tools have something to act on.

Connect your AI client

OrionBI is a standard remote MCP server. In your client, add it as a custom / remote MCP server using this URL:

https://orionbi.ai/mcp

Authentication is OAuth 2.1 with PKCE and supports dynamic client registration, so most clients register themselves automatically — there are no API keys or client secrets to paste. You sign in through your OrionBI identity provider, and the client never sees your password. Menu names below can drift between client releases; if a label has changed, look for the equivalent Connectors or MCP area.

Claude (web and desktop)

Open Settings → Connectors (sometimes shown as Customize → Connectors) and choose Add custom connector (or find OrionBI in the connectors directory once it is listed).
Enter the server URL https://orionbi.ai/mcp and click Add — leave the OAuth fields blank; dynamic registration handles them.
Click Connect and complete the OAuth sign-in in your browser.
In a chat, open the + (connectors) menu and enable OrionBI so its tools are available.

On Team or Enterprise plans, an owner adds the connector once under Organization settings → Connectors; each member then connects and authenticates individually. Free plans allow a single custom connector.

ChatGPT

Open Settings → Plugins and select the MCPs tab.
Click Add server.
Set Type to Streamable HTTP.
Enter a Name (for example orionbi) and set the URL to https://orionbi.ai/mcp. Leave the bearer-token and header fields empty — OrionBI authenticates over OAuth, so no token is needed.
Click Save, then complete the OAuth sign-in when prompted.
Make sure the server is toggled on in the MCPs list, then call its tools from a chat.

Cursor

Open Settings → Tools & MCP and click New MCP Server — this opens your mcp.json (global ~/.cursor/mcp.json or project .cursor/mcp.json). You can also edit that file directly.
Add the server with just its url — no type and no auth block, because dynamic client registration means there is no client ID or secret to add:
```
{
  "mcpServers": {
    "orionbi": {
      "url": "https://orionbi.ai/mcp"
    }
  }
}
```
Save the file. Cursor connects and opens a browser OAuth sign-in; complete it and approve access.
Back in Settings → Tools & MCP, confirm the server shows as connected and enable its tools if any are toggled off, then call them from Agent chat.

Any other MCP client (VS Code, Windsurf, Zed, and more)

Because OrionBI is a standard remote, Streamable-HTTP MCP server with OAuth, it works with any client that supports remote MCP servers. Wherever your client lets you add one — a settings panel or an mcp.json-style config — add https://orionbi.ai/mcp as an HTTP / remote server (type http) and complete the OAuth prompt. See your client's own MCP documentation for the exact location.

Once connected, the tools available to you are mounted based on what an administrator has enabled for your account.

Available tools

Databases

postgres_list_connections — list the Data Sources you can use, including SaaS sources (Google Sheets, Airtable, Salesforce, HubSpot), which are marked as usable only through a Data Set.
postgres_schema — browse schemas, tables, and columns.
postgres_query — run a strictly read-only query in the engine's native language. It executes inside a read-only transaction, so any mutating statement is rejected.
postgres_execute — run a single write statement (INSERT/UPDATE/DELETE or DDL) against a connection you own that has Allow writes enabled. It is flagged as a write, so clients that support confirmations prompt you before it runs.

Supported engines include PostgreSQL, MySQL, SQL Server, Oracle, Snowflake, BigQuery, Redshift, Databricks, ClickHouse, MongoDB, Cassandra, Redis, Elasticsearch, DynamoDB, and Neo4j.

Data Sets

dataset_list / dataset_get — list cached data sets and inspect their columns and freshness.
dataset_query — run read-only SQL against materialized data sets built from sources like Google Sheets, Airtable, Salesforce, and HubSpot.
dataset_create, dataset_update, dataset_refresh — materialize a query (or a SaaS source) from a Data Source you own, change its SQL, schedule or refresh mode, and rebuild it on demand.

GitHub

Read: github_list_repositories, github_list_issues, github_get_issue, github_list_pull_requests, github_get_file_contents, github_search_repositories, github_search_code.
Write (on connections you own): github_create_issue, github_create_issue_comment.

Dashboards

dashboard_publish, dashboard_check, dashboard_patch, dashboard_get, dashboard_list, dashboard_versions, dashboard_rollback, dashboard_promote_version, dashboard_share, dashboard_reference — publish a BI dashboard from a JSON spec, hosted on OrionBI and read live from your Data Sources and Data Sets; check it before publishing; change part of it; keep every version; invite a person to view or edit it.
dashboard_rename, dashboard_share_list, dashboard_unshare — rename a dashboard without changing its address, see who it is shared with, and remove a person from it.
dashboard_read — read the figures any dashboard you can open shows, as its page shows them. The queries behind it are never returned.
The spec — every panel kind, format, filter, layout and theme field — is documented at Dashboards reference, the same text dashboard_reference returns.

Session

tools_list_available, tools_attach, tools_detach — see which tools are available and attach or detach them for your session.

Setting up resources

Resources are configured in the OrionBI admin UI, not in Claude:

Databases — add a connection under Connections with its host, credentials, and SSL. Turn on Allow writes only if you want postgres_execute to run mutating statements against it.
Data Sets — build a materialized data set from a database or SaaS source under Data Sets and set a refresh schedule.
GitHub — connect a GitHub account under Tools → GitHub using GitHub's device flow. Your token is encrypted at rest and never exposed to your AI client.

Reads, writes, and confirmations

Read-only tools run without a per-call confirmation. Any tool that writes — postgres_execute, github_create_issue, and github_create_issue_comment — is marked as a write, so your client can prompt you before it runs. Writes are always owner-only: a resource shared with you is read-only and can never be modified through your session.

Governance, sharing, and audit

Administrators control exactly which tools each user or group can use. Connections and data sets can be shared read-only with teammates, and every action — including every write — is recorded in an immutable audit log.

Support

Questions or issues? Email helpdesk@orionomega.dev.

Connecting OrionBI to your AI client