Your connected sources

Every source connected in the workspaces you granted your agent is reachable through Vistazo's MCP server. MCP is the open standard your agent uses to reach Vistazo. Your agent reaches a database or app connected in Vistazo without adding another server or holding any credentials. Vistazo runs each call with the connection's stored credentials.

Four tools on the server carry every source:

1. search_tools finds a tool by keywords, by source, or both. Each match names the tool that runs it.
2. describe_tool returns that tool's arguments.
3. call_tool runs a tool that only reads.
4. call_write_tool runs a tool that changes something. Your client asks you before each call.

Each source tool is named <source>__<tool>, such as economy__series_observations or salesforce__list_leads. The /mcp/full address also lists every source tool under that name, for clients that load tools only when they need them (Connect your own agent). A call by that name runs on every address.

search_tools finds a source the moment you connect it. A client on /mcp/full keeps the tool list from when it last loaded it, so reload the list or restart the client to see the new tools there.

Every call runs in a workspace you ticked on the consent screen, and the server refuses a call that names any other, whatever the client sends. get_me takes no arguments and lists your workspaces, marking each one this connection cannot reach as reachable: false. Each reachable workspace also lists its connected sources, with how many tools each holds. Every other core tool takes a workspaceId, apart from search_tools and describe_tool, and the server never assumes one.

Each source tool, and each dataset tool, takes an optional workspaceId. With one workspace granted, your agent can leave it out. With several, a call that leaves it out runs in the workspace that has the tool, if only one does. If more than one has it, the call is refused and the reply lists the workspaces to choose from.

A tool connected in only some of your workspaces says in its description where it is available.

The permission set on each tool under Settings → Connections decides what your agent can call. Connections covers how to set it.

Your client asks you before every call_write_tool call, even for a tool set to Allowed, because that one tool runs every write. On /mcp/full a source tool called by its own name carries its own marking instead, so an allowed write runs without the prompt.

The hosted sources offer only their read-only tools. A workspace where you are a guest adds no source tools, and a call into it is refused. The same holds where your license lets you view but not build.

A source tool is for asking a source directly. A widget's pipeline names the connection and its tool separately, as mcpId and toolName, and update_widget's pipeline refuses a <source>__<tool> name.

A source tool returns the source's own result. A result over 100,000 characters is cut, with a note asking your agent to narrow the arguments.

Each MCP connection, one client signed in once, may run 300 data fetches an hour. Source calls, the refreshes your agent starts and the test run of a new pipeline all count toward it.

The five public sources Vistazo hosts are reached the same way, through search_tools and call_tool. They have no address of their own to add.

The tools that read your uploaded files are core Vistazo tools, documented under Datasets. Calling a source tool is free, the hosted sources included. Metering has the details.