Contextflo MCP¶
A governed context layer for team analytics in chat. Contextflo connects your data warehouse to the AI tools your team already uses: hook up BigQuery, Snowflake, Postgres, Redshift, Databricks or ClickHouse (or upload a CSV), then ask questions of the real data and build live, auto-refreshing dashboards by chatting with Claude or ChatGPT over MCP. Table-level access control keeps agent access scoped.
Server type: Hosted (Streamable HTTP)
Endpoint: https://mcp.contextflo.com/mcp (anonymous initialize returns 401 Authentication required - live)
Auth: Contextflo account sign-in (OAuth flow; see the MCP client guides)
Sources: BigQuery, Snowflake, Postgres, Redshift, Databricks, ClickHouse, CSV upload
Access control: table-level permissions
Outputs: chat answers, live auto-refreshing dashboards
Docs: contextflo.com/docs (MCP client guides for Claude and ChatGPT)
Built by: Contextflo (contextflo.com)
Why This Matters for Operators¶
Self-serve analytics has been stuck between "give everyone a BI tool nobody logs into" and "answer every question in SQL yourself". Contextflo's MCP picks a third lane.
First, questions in chat, against real data. The agent queries the connected sources directly - "what was last week's churn by plan tier" - instead of waiting for a dashboard request to land in the data team's queue.
Second, governance is built in, not bolted on. Table-level access control means the whole team can query without handing the agent a warehouse superuser. That is the difference between a demo and something you can actually roll out.
Third, dashboards become a chat product. Per the vendor's blog, live auto-refreshing dashboards get built by chatting - no Tableau, no drag and drop - and stay connected to the sources.
Tools and Capabilities¶
Capability-level table from the vendor's docs; exact tool names require sign-in - the endpoint refuses anonymous enumeration (initialize returns a JSON-RPC 401 "Authentication required", which confirms liveness).
| Capability | Description |
|---|---|
| Ask data questions | Natural-language queries against connected warehouses and CSV uploads |
| Build dashboards | Live, auto-refreshing dashboards generated from chat, connected to the sources |
| Schema and table discovery | Work with the real table and schema names of your sources |
| Permission-scoped execution | Queries run against the table-level permissions granted to the workspace |
Verification (Aug 30, 2026)¶
- Endpoint live-verified: anonymous initialize at mcp.contextflo.com/mcp returns JSON-RPC error -32001 "Authentication required" (liveness with sign-in auth, same class as the Taskfolk and Jitsu verifications)
- Vendor docs: contextflo.com/docs carries MCP client pages for Claude and ChatGPT naming the endpoint
- mcp.so listing (Jul 9, 2026) and vendor blog confirm the product surface; GitHub org exists with 0-star repos and no public MCP repo
Notes and Caveats¶
- Hosted only: no self-host or stdio variant
- Young vendor: no public MCP repo, all public org repos are 0-star; treat as early adopter
- Sign-in required for everything past the endpoint check
- Warehouse credentials and table grants live in the Contextflo workspace - review their security model before connecting production data