Studio MCP

The Studio MCP lets AI assistants like Claude and ChatGPT connect to your live Evidence workspace. Once connected, the assistant can access your projects, published reports, and query your data. The server is available at https://evidence.studio/api/mcp and uses Streamable HTTP.

Every request runs as the signed-in user — access policies and page controls are respected.

Adding to AI Tools

Claude (Web & Desktop)
ChatGPT
Claude Code
Cursor

Evidence Studio is available as a connector in the Claude directory:

Go to Claude.ai → Customize → Connectors
Click Add → Browse connectors, then search for “evidence-studio”, or use this direct link
Click Connect / Add to my Team
Authenticate with Evidence Studio — you’ll sign in and choose the organization to authorize

The Evidence tools will now be available in your Claude conversations.

claude mcp add evidence-studio --transport http https://evidence.studio/api/mcp

Then run /mcp inside Claude Code and authenticate — you’ll be sent to Evidence to sign in and choose the organization to authorize.

Add the following to your Cursor MCP configuration (.cursor/mcp.json):

{
  "mcpServers": {
    "evidence-studio": {
      "url": "https://evidence.studio/api/mcp"
    }
  }
}

Open Cursor → Settings → MCP and click Login on the evidence-studio server to sign in and choose the organization to authorize.

Available Tools

The Studio MCP server provides the following tools for AI assistants:

Tool	Description
`whoami`	Get the authenticated Evidence user and their active organization.
`list_organizations`	List the Evidence organizations you belong to. The active organization is fixed to the one you authorized; to act in a different organization, reconnect and select it during sign-in.
`list_projects`	List the projects (workspaces) in this organization that you have access to.
`get_context`	Get the workspace context (the projects’ agent/context instructions and available skills) that Evidence maintains for AI assistants. Call with no arguments to get context for every project you can access, or pass a projectSlug to scope it to one project. Read this before answering questions about the workspace’s data, metrics, or conventions.
`list_project_pages`	List the pages within a specific project that you have access to. Use after list_projects.
`search_published_content`	Search published reports across the workspace by free text (titles + page body). Prefer this when the user asks about a topic, metric, or term.
`get_page_summary`	Get a structured summary of a published page (title, queries, components) so you can describe what it shows.
`get_page_content`	Get the raw Markdoc content of a published page. Prefer get_page_summary first.
`list_project_files`	List the reusable source files that report pages build on: partials, saved SQL query files, and custom components. Their content lives in separate files that get_page_content does NOT expand — use this to discover them, then get_project_file to read one. For report pages themselves, use list_project_pages.
`get_project_file`	Read the raw content of one source file a page references — a partial, a SQL query file, or a custom component — by its path. This is how you follow a `{% partial file="x" /%}` reference or inspect the SQL behind a chart; get_page_content returns pages WITHOUT expanding these. Pass the path as referenced (e.g. “l10/cumulativemaus”). The same path can name both a partial and a query — if so, pass `type` to pick which one. For report pages, use get_page_content.
`list_tables`	List the warehouse tables and models available in this workspace. Pass `filter` to narrow using a substring.
`get_metadata`	Get column names and types for the given tables so you can write correct SQL.
`run_query`	Execute a read-only SQL query against the workspace data you have access to. Before running your first query, call get_context. Row-level security is applied based on who is asking. SELECT/WITH only; default LIMIT 100, max 1000. Do not add SETTINGS or FORMAT clauses, these are added by server.
`search_docs`	Search the Evidence documentation (components, core concepts, features, data sources, guides). Use this to check component syntax before calling visualize, and after a validation error.
`read_doc`	Get the full markdown of an Evidence documentation page. Component pages carry the full attribute list and a worked example.
`visualize`	Render an Evidence Markdoc snippet to an image — a chart, table, big value, or any mix of components and prose. Provide the full markdown including any ```sql inline query blocks it references. Use existing markdown if available.

run_query is only available when your organization allows query execution. If it’s disabled, the assistant can still explore tables, metadata, pages, and search results.

Get Started

Core Concepts

Features

MCP

CLI

Self Host

Direct Connectors

Evidence Warehouse

Components

Adding to AI Tools

Available Tools

​Adding to AI Tools

​Available Tools

Adding to AI Tools

Available Tools