Ability to request/require agents to utilize mcp servers for fetching or saving context

[Shahab96]

Right now the way I do this is by just adding information into the instructions directory for specific commands, but I think there may be a better way of doing this. I'm more than happy to work on this but I wanted to make sure that I'm not the only one who wants this.

Example:

In the create-tasks.md file

... steps to be performed

<step number="x" name="jira sync">
  Use the atlassian mcp server to create these tasks in jira and assign them the appropriate labels. Be sure to create them with the proper hierarchy so that dependent tasks are marked as such.
  etc etc
</step>

I think there may be a better way of doing this, such as by defining specific files for mcp servers and how the user wishes to integrate them into their workflow and then they may be referenced with some particular instructions, similar to what we see in the code-style.md file for things like:-

<conditional-block task-condition="sql" context-check="sql-style">
IF current task involves writing or updating sql:
  IF sql-style.md already in context:
    SKIP: Re-reading this file
    NOTE: "Using Sql style guide already in context"
  ELSE:
    <context_fetcher_strategy>
      IF current agent is Claude Code AND context-fetcher agent exists:
        USE: @agent:context-fetcher
        REQUEST: "Get Sql style rules from code-style/sql-style.md"
        PROCESS: Returned style rules
      ELSE:
        READ: @.agent-os/standards/code-style/sql-style.md
    </context_fetcher_strategy>
ELSE:
  SKIP: Sql style guide not relevant to current task
</conditional-block>

There may be a way for us to create such logic in the create-spec command for example, which uses a similar syntax to above and references mcp servers similar to how we reference agents in claude code.

I'd also love to see how other people envision doing this, my own idea here is just something I came up with on the fly and there's probably much better ways to do this.

[coryt]

I took a similar approach for linear but I just track the work at the spec level. I left the individual tasks defined in the spec and track what I'm working on and future specs in linear.