May I know why we decide not to expose the other tools for more granular control of actors?

[kuoyusheng]

Curious to understand why we only expose few tools now. It makes me hard to check , review the run I have. Is this a resource concern? is it possible to bring them back before we have dataset in resources? the data response right now is very limited.

[MQ37]

@kuoyusheng Hey, we decided to temporarily disable all the tools except for Apify documentation, Actor calling and discovery (search, dynamic adding) since in some cases it confused the LLM and sometimes started calling tools totally randomly while trying to "troubleshoot" a failed Actor run (that failed because the client does not support the tools list changed notification). Right now, we are focusing on writing evals to ensure the Apify MCP server works with most of the LLMs and clients, and then we will re-introduce these tools once the actual workflows and descriptions of the tools are polished.

[jirispilka]

We will add a new switch:

--tools=runs

options: runs,storage,builds,tasks,logs,docs
default: runs (when we have proper evals and everything works we will make default=all)

Context private-slack

[kuoyusheng]

so this --tools option can only be done in stdio version not remote correct? or there are work around.

[MQ37]

so this --tools option can only be done in stdio version not remote correct? or there are work around.

It will work both for the stdio and remote (mcp.apify.com) versions.

stdio - CLI args (for example --tools runs,docs)
remote - URL query params (for example https://mcp.apify.com?tools=runs,docs)

There is a PR draft for that. We just need to decide what tools will be enabled by default for now - for sure docs, and maybe runs. Once we test and polish everything, we can enable all.

[MQ37]

Hey @kuoyusheng, we just released a new version 0.2.14 of @apify/actors-mcp-server that features the --tools switch to enable a specific category of tools.

For example, to include runs and storage tools to be able to work with Apify storage and get info and logs from Actor runs, run the Actors MCP server like this: npx -y @apify/actors-mcp-server@0.2.14 --tools runs,storage

Tool categories are as follows:
- docs: Search and fetch Apify documentation tools.
- runs: Get Actor runs list, run details, and logs from a specific Actor run.
- storage: Access datasets, key-value stores, and their records.
- preview: Experimental tools in preview mode.

Note: Tools that enable you to search Actors from the Apify Store and get their details are always enabled by default

This switch will also work on the remote https://mcp.apify.com MCP server once we do a release there (should be in couple of days), using the same input but in the URL query params like https://mcp.apify.com?tools=runs,storage.

[jirispilka]

@MQ37 The issue should only be closed after we do a release—we shouldn’t expect users to keep track of mcp.apify.com releases themselves.

Sorry, I missed that in the review. We also need to update the documentation in Actor.md, Readme.md, and on docs.apify.

[MQ37]

@MQ37 The issue should only be closed after we do a release—we shouldn’t expect users to keep track of mcp.apify.com releases themselves.

Sorry, I missed that in the review. We also need to update the documentation in Actor.md, Readme.md, and on docs.apify.

Agree with the release, we will close this after the mcp.apify.com release.

Regarding the docs, I will update them in a separate PR 👍

[kuoyusheng]

hi @MQ37 super thanks for the updates. will try out the new version.

[MQ37]

Hey, @kuoyusheng the new version is live on mcp.apify.com - for more details visit https://docs.apify.com/platform/integrations/mcp#dynamic-actor-tooling. Closing this issue, feel free to reopen.

FYI @jirispilka