docs: MCP Server #7360
docs: MCP Server #7360
Conversation
|
@royendo FYI we now support MCP both for Rill Developer and Rill Cloud, and the setup has become a bit simpler. More details here: #7357
|
|
@mindspank , plan to release this tomorrow morning with the blog, etc. LMK any changes needed :) |
docs/docs/explore/mcp.md
Outdated
| @@ -0,0 +1,158 @@ | |||
| --- | |||
| title: "Rill MCP Server" | |||
| description: How to connect to Rill MCP and query your deployment | |||
There was a problem hiding this comment.
query your data or query your metrics views might be more appropriate
docs/docs/explore/mcp.md
Outdated
| ### Create a Rill service token (if your project is on Rill Cloud) | ||
| ``` | ||
| # Install the Rill CLI if you haven't already | ||
| curl https://rill.sh | sh | ||
|
|
||
| # Create a service token | ||
| rill service create mcp-server | ||
| ``` |
There was a problem hiding this comment.
We should recommend user tokens, not service tokens. See: #7359
There was a problem hiding this comment.
:done:, you think we'll have the UI in time to put a screenshot here and steps to get creds in the UI?
There was a problem hiding this comment.
Yes! Polishing it up now
|
Should there also be docs on installing mcp-remote package? It only says you need nodejs which is a bit barebones |
You shouldn't need to install it, npx should auto install it however we could add a -y flag to ensure confirmation on install is accepted per automatic. @ericpgreen2 see any issues with that? |
docs/docs/explore/mcp.md
Outdated
|
|
||
| The Rill Model Context Protocol (MCP) server exposes Rill's most essential APIs to LLMs. It's designed primarily for data analysts, not data engineers, and focuses on consuming Rill metric views—not creating them. | ||
|
|
||
| The server can also generate VegaLite-backed visualizations from natural language prompts and data tables. However, not all MCP clients support the Image datatype. |
There was a problem hiding this comment.
@ericpgreen2 is this still the case?
There was a problem hiding this comment.
No, good catch, it doesn't generate visualizations currently.
No issue with that. It looks like that's a common troubleshooting step that |
|
What part of the yaml is retrieved for the llm to help decide which metrics view to use? Is it the "description" for the metricsviews to decide the right metrics view and the "description" field on each dim and measure to decide what to query? Referencing that information encourages folks to better improve their docs to get better results from their llm: https://docs.rilldata.com/reference/project-files/metrics-views If descriptions are pulled this would also help us tell the LLM which metrics views to NOT use (in the yaml description) as they may be prototypes not suitable for reporting, and thus get better results out of the box. 
It seems to only have access to the name of the metricsview and dims/measures which is something that should be improved soon: Line 125 in fa9f7c0 |
|
Because v63 is live, we need to publish this docs page so we don't have a broken link in the application. I'll approve & merge, and we can take follow-ups in a separate PR. |
INSERT DESCRIPTION HERE
Checklist: