MCP Server
The Querri MCP server lets AI assistants like Claude, ChatGPT, Cursor, Windsurf, and VS Code Copilot talk to your Querri data on your behalf. Once connected, you can ask questions about your data, run multi-step analyses, and reference dashboards — all from inside the AI tool you already use.
What is MCP?
Section titled “What is MCP?”MCP (Model Context Protocol) is an open standard that lets AI assistants connect to external systems — databases, SaaS tools, internal APIs — through a single, consistent interface. You install or enable an MCP “server” inside your AI tool, log in once, and the assistant gains a set of tools it can call when you ask questions.
For Querri, the MCP server gives your AI assistant the same powers you have: it can list your data sources, run analyses, summarize results, and link to dashboards. You stay in your favorite AI tool; it pulls data from Querri as it needs to.
What you can do
Section titled “What you can do”After connecting, you can prompt your AI assistant naturally. It picks the right tool automatically. Examples:
- “What sources are available in Querri?” — lists your data sources
- “Look at our Q3 sales data and tell me which regions are growing fastest.” — kicks off a Querri analysis and returns the answer
- “Continue the analysis we started last time and break it out by product line.” — picks up an existing project conversation
- “Show me the dashboard structure for our revenue dashboard.” — pulls dashboard widgets and links
- “Give me the raw step output from that join we ran earlier.” — fetches the rows from a specific analysis step
The assistant decides when to call Querri. You don’t need to remember tool names — just ask.
Connecting your AI assistant
Section titled “Connecting your AI assistant”The Querri MCP server is hosted at:
https://app.querri.com/mcpIt uses OAuth authentication: when you connect, you’ll be redirected to Querri’s normal SSO login. Once approved, your assistant has access scoped to whatever you have access to in Querri.
Claude (claude.ai or Claude Desktop)
Section titled “Claude (claude.ai or Claude Desktop)”- Open Settings → Connectors
- Click Add custom connector
- Enter the MCP server URL:
https://app.querri.com/mcp - Complete the SSO login when prompted
- Start a new conversation — Querri tools appear automatically when relevant
ChatGPT
Section titled “ChatGPT”- Open Settings → Connectors → Advanced
- Add a new MCP server with URL
https://app.querri.com/mcp - Complete the OAuth flow
- Mention Querri in your prompt to invoke the tools
Cursor, Windsurf, VS Code Copilot, and other MCP clients
Section titled “Cursor, Windsurf, VS Code Copilot, and other MCP clients”Most MCP-compatible IDEs and assistants accept a remote MCP server URL in their settings. Add https://app.querri.com/mcp as a remote MCP server. The exact menu varies — look for Settings → MCP, Extensions → MCP servers, or similar.
For tools that read a configuration file, point the entry at the URL above and choose OAuth as the auth method.
Authentication
Section titled “Authentication”Authentication uses Querri’s standard SSO. The first time you connect, your assistant opens a browser window where you log in normally. The assistant stores a refresh token so you don’t have to log in again on subsequent sessions.
Your AI assistant inherits your role and permissions:
- Members can browse and analyze data they have access to.
- Admins can additionally see organization-wide projects and dashboards.
Row-level security and access policies are enforced exactly as they are in the web app — your assistant cannot see anything you can’t.
If you belong to multiple organizations, the assistant uses the org tied to your current login. To switch, you’ll need to log out of Querri and log back in to the target org, then reconnect the assistant.
Available tools
Section titled “Available tools”The Querri MCP server exposes these tools to your AI assistant:
Asking questions
Section titled “Asking questions”| Tool | What it does |
|---|---|
ask_project | Kick off a multi-step analysis on your data. The default for any analytical question — handles joins, filtering, grouping, figures. Querri’s planner discovers the right sources automatically. Returns a chat ID; the assistant then polls get_chat for the answer. |
ask_source | Shortcut for questions clearly answerable from a single, named data source with no joins needed. Returns the answer directly. |
get_chat | Reads back the latest answer in a project chat. Used by ask_project to wait for results, and also for catching up on a project’s prior conversation. |
Browsing your workspace
Section titled “Browsing your workspace”| Tool | What it does |
|---|---|
list_sources | Lists data sources with column counts and row counts. |
get_source | Returns a source’s column schema (names and types). |
list_projects | Lists analysis projects you have access to. |
get_project | Returns a project as a Mermaid diagram of its analysis steps, plus a step-ID table. |
list_project_steps | Lists the steps in a project. |
list_views | Lists SQL views defined in your organization. |
get_view | Returns a view’s SQL definition and metadata. |
list_dashboards | Lists dashboards. |
get_dashboard | Returns dashboard metadata, widgets, and links. |
list_files | Lists uploaded files. |
Working with raw data
Section titled “Working with raw data”| Tool | What it does |
|---|---|
query_source | Runs a SQL query against a single source. The source is exposed as a view named data. |
get_step_data | Returns the rows produced by a specific step inside a project. Useful when you want to inspect intermediate output. |
Workspace management
Section titled “Workspace management”| Tool | What it does |
|---|---|
create_project | Creates an empty project. Most of the time ask_project is what you want — it creates the project and runs the analysis in one call. |
list_organizations | Lists organizations you belong to. |
switch_organization | Returns instructions for switching to a different org (requires re-authentication). |
Tips for good results
Section titled “Tips for good results”- Be specific about the question, not the steps. Say “compare year-over-year revenue by region” rather than “join sales and regions, then group by region.” Querri’s planner picks the steps.
- Reference projects to continue analysis. If your assistant ran an analysis earlier, mentioning it (or letting it pass
project_id) keeps the same project’s context. - Long-running analyses are normal. Multi-step analyses take 30 seconds to a few minutes. The assistant will wait and report when done.
- Ask for figures explicitly. “Plot revenue over time as a line chart” — Querri returns the figure inline.
- Use SQL for known-shape questions. When you already know which source and just need to slice it a particular way,
query_sourceis faster thanask_project.
Troubleshooting
Section titled “Troubleshooting”The assistant says it can’t connect to Querri. Re-run the OAuth flow from your assistant’s connector settings. Tokens expire and need to be refreshed via login.
Tools don’t appear in conversation. Some assistants only surface MCP tools when the prompt clearly relates to the connector. Try mentioning Querri or your data explicitly: “Use Querri to look at…”
An analysis seems stuck. Long analyses long-poll get_chat for up to ~25 seconds at a time. The assistant should be calling it repeatedly. If a single analysis takes more than ~5 minutes, ask for a simpler version of the question.
Wrong organization showing. Log out of Querri, log back in to the right org, then reconnect the MCP server.
Just need a Python library?
Section titled “Just need a Python library?”If you’re building a Python application that talks to Querri directly — without going through an AI assistant — use the Python SDK instead. The SDK gives you typed sync and async clients for the same data.
To use Querri from a terminal or shell script, see the CLI.