mcp: initial sampling implementation #249946
Merged
Conversation
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
- MCP servers get a prompt when they ask for sampling. Usage in tool calls (i.e. a sampling request within a concurrent tool request) is a separate prompt from 'ambient' usage. Sampling is 'just' a request to the client and can technically happen at any time. - By default MCP servers will get the default model in chat. But that is configurable by users. Multiple models can be configured and if they are then the model will be picked based on the rough 'model hints' we get from the server. - Not sure where the config should live. I have it in user settings in this demo. Having it in mcp.json would be a natural place to put it, but that doesn't work for extension-provided servers or servers from other config files (e.g. claude desktop configs) so that could not be the only place. - There are some limitations from MCP such as modelcontextprotocol/modelcontextprotocol#91
9 tasks
connor4312
commented
May 28, 2025
|
|
||
| const model = await this._getMatchingModel(opts); | ||
| // todo@connor4312: nullExtensionDescription.identifier -> undefined with API update | ||
| const response = await this._languageModelsService.sendChatRequest(model, new ExtensionIdentifier('Github.copilot-chat'), messages, {}, token); |
There was a problem hiding this comment.
This is a 🩸, but to keep this PR focused I'll open a second PR for the necessary API change. Sampling is rarely used yet today so I do not expect any big issues.
aeschli
approved these changes
May 28, 2025
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
calls (i.e. a sampling request within a concurrent tool request) is a
separate prompt from 'ambient' usage. Sampling is 'just' a request to
the client and can technically happen at any time.
configurable by users. Multiple models can be configured and if they
are then the model will be picked based on the rough 'model hints' we
get from the server.
this demo. Having it in mcp.json would be a natural place to put it,
but that doesn't work for extension-provided servers or servers from
other config files (e.g. claude desktop configs) so that could not
be the only place.