> ## Documentation Index > Fetch the complete documentation index at: https://docs.vistazo.ai/llms.txt > Use this file to discover all available pages before exploring further. # Vistazo as an MCP server > Build and manage dashboards from your own agent. Vistazo is an MCP client — it reads your data sources over MCP. It is also an MCP **server**, so your own agent can drive it. Connect it to Claude Code, Cursor, Codex, Claude Desktop, Gemini CLI, VS Code, or anything else that speaks remote MCP, and your agent can build dashboards, wire widgets to data, manage parameters and schedules, read what a dashboard currently says, and administer a workspace. ``` https://mcp.vistazo.ai/mcp ``` Copy-paste setup for the common clients, plus what the consent screen is asking. ## What is on the surface Around 80 tools, grouped as they are documented. Resolve who you are and which workspaces you can reach. Create, find, describe, clone, version. Add, wire to data, style, lay out. Filters and date ranges. Re-fetch, check freshness, explore live values. Links, access requests, ownership. The data sources a workspace can read. Tables from uploaded files. Read and write workspace procedures. Custom chart code. Members, seats, limits, audit log. Asset ownership and identity. Admin and governance tools are only registered for a caller who is an admin somewhere, so most clients load a smaller surface than the full list. ## Conventions worth knowing before you start **Workspace is always explicit.** Every tool takes a `workspaceId`, apart from `get_me` and `list_workspaces`, which take no arguments at all. Nothing is inferred from context, and a workspace outside what you granted this client is refused. Call `list_workspaces` first, and skip any entry marked `reachable: false`. **Search to find, list to browse.** `search_dashboards` matches names, tags and descriptions and ranks hits, while `list_dashboards` is for enumerating and triaging. Listing everything and filtering the names yourself is the slow path. **Read cached data before refreshing anything.** `get_widget_data` returns the last published rows with no fetch and no cost. `get_dashboard` returns a written summary of what the dashboard means, including a plain-language reading of each widget, which is usually enough to answer a question without touching a source. Refresh when the user needs data fresher than the last run, not as a reflex. **A widget is a pipeline plus a payload.** The pipeline names data-source tools and a transform; the payload is the chart config. `set_widget_pipeline` replaces the whole pipeline; `patch_widget_pipeline` adjusts arguments. Transforms are validated before any write, so a bad one is refused rather than leaving a broken widget. **Annotations are honest.** Read-only tools say so, and a write that has not explicitly declared itself additive is marked destructive, following the MCP spec's default. Your client's confirmation prompts land in the right places as a result. ## Curated data sources The five public data sources Vistazo hosts are separate MCP endpoints, so your agent can read them directly without going through a dashboard: Development indicators. Conditions, forecasts, history. Public company financials. Federal fiscal data. Daily currency rates. Tools that read your own uploaded tables are on the core surface, so reading your own files needs no second connection. See [Datasets](/mcp-tools/datasets). ## Metering Calls to Vistazo's **data** tools are metered against your allowance and appear under **Tools** in your usage breakdown: the curated sources, and reading your own datasets. The dashboard and widget tools are not billed per call. What they trigger is — a refresh, or the model run that authors a transform. Rate limits, which count every request rather than only tool calls, are in [Limits](/reference/limits).