Your data doesn’t all live in files you upload. Some of it sits in a database your team runs, some behind an MCP server (a small service that gives an AI assistant a set of tools, such as “look up a customer”), and some behind an API (a web address that hands back data). On the Connections page you connect these once, and Crunch can then use them in a chat, with a read-only database table read where it lives and nothing copied.
What you’ll need
- Sign-in details for what you’re connecting: a database user and password, an API key, or a bearer token (a long secret string that says who you are). Use a read-only user for a database so Crunch can read your tables but never change them.
- Your organization has to allow the kind you’re adding. Where it doesn’t, the page says so, for example “Your organization doesn’t allow MCP servers.” or “Your organization doesn’t allow API connections.”
What’s on the page
Open Connections under Data and connections in the left sidebar. It has these sections:
- Connections. Everything you’ve connected, each with its kind, its address and the credential it uses. Next to Add are three buttons: A database, MCP server from a credential and An API.
- Linked tables. Database tables Crunch reads in place.
- Credentials. The sign-ins your connections use.
- On this computer. MCP servers on your own computer, shown only in Crunch Desktop (below).
Add a credential first
A secret is typed once and never shown again, so credentials are kept apart from the connections that use them. That way you can replace a password without rebuilding the connection.
- In Credentials, click Add a credential.
- Choose the kind: Password, API key or Bearer token.
- Give it a Name, fill in User (for a password) or an optional Label, and type the secret in the Password, Key or Token field.
- Click Save credential.
Once it’s saved, the secret is never shown again, by Crunch or by you. To change it, click Replace on the credential and type the new one. Delete removes it, and warns you which connections will stop working.
Add a database
- In Connections, click A database.
- Choose the Kind: Postgres, MySQL or Iceberg REST catalog (a way data lakes list their tables).
- Give it a Name. For Postgres and MySQL, fill in Host, Port, Database and, for Postgres, optionally Schemas, then pick the Credential with its password. For a catalog, fill in Catalog URI and Warehouse, choose how it signs in under Sign in with (Bearer token or Nothing), and optionally list Namespaces.
- Click Add connection.
Link a table
A database connection holds nothing to ask about until you link a table from it.
- Under Linked tables, click Link a table.
- Choose the Database connection and type the table’s Address:
schema.tablefor Postgres,database.tablefor MySQL andnamespace.tablefor a catalog. - Click Preview to see its columns and first rows.
- Name the dataset under Dataset name and click Link.
The table is then listed under Linked tables and can be picked in any chat like other data. See Datasets.
Add an MCP server
- Click MCP server from a credential.
- Fill in a Name and the server’s URL, and choose a Credential, or None if it needs no sign-in.
- Click Connect. Crunch asks the server what it offers.
- For each tool, tick Let through to allow it in chats. Reads only marks a tool that only looks things up. Click Save tools.
Only the tools you let through are available. Click Tools on the connection later to change them. If your organization allows only tools that read, the page says so, and tools that act aren’t offered in chats.
Add an API
- Click An API.
- Fill in a Name and the Base URL. OpenAPI URL (optional) is the document that describes
the API; if you leave it empty, Crunch looks for
/openapi.jsonon the base URL. Choose a Credential. If it’s an API key, you can set the API key header (optional), such asX-API-Key. - Click Create, choose the operations to Let through, and click Save tools.
API tools are read-only: Crunch only makes GET requests, so it can look data up but not change anything. If the API lists no GET operations, the page says so.
MCP servers on this computer
If you use Crunch Desktop, the On this computer section lists MCP servers from Crunch Desktop’s MCP settings file on your computer. Click Find MCP servers, then Choose tools on each to pick which tools to let through. Forget removes it. A chat uses these servers only on that computer, because that’s the only place they run. See Crunch Desktop.
Use a connection in a chat
In a message box, click + and choose a connection under Connections, or type
/connection- and its name. It is used from your next message. You can also add connections to a
workspace so every chat there has them.
When someone shares a workspace with you
If a workspace shared with you uses connections that need a sign-in, open the workspace and look under Connections of this workspace. Click Connect your account, pick one of your own credentials and click Connect, so the connection runs as you. If the owner chose to let you use their sign-ins, you don’t need to; you never see their credentials either way. Disconnect removes yours.
How to tell it worked
The connection is listed under Connections, with the address and credential it uses and, for MCP servers and APIs, a line such as “3 tools let through”. A linked table appears under Linked tables.
Good to know
- Heads up: deleting a connection asks first. Deleting a credential stops every connection that uses it.
- A secret you’ve saved can’t be read back, so keep your own copy somewhere safe.
- A “No tools let through yet.” line means the server is connected but nothing is allowed yet. Click Tools and let some through.
Related
- Workspaces: keep connections together with the data and notes for one job.
- Semantic notes: tell Crunch what a connection’s data means.
- Datasets: pick a linked table for a conversation.
- On-prem sync connector: for a database behind your firewall.