Repository navigation
How will tool visibility be controlled after tools config is removed? #18051
Description
Question
Hi,
I’m trying to understand the intended migration path from the deprecated tools configuration to the newer permission system.
From the documentation, it seems that:
tools(boolean enable/disable) is deprecatedpermissionis now the recommended configuration
However, these two mechanisms appear to control different things:
tools:
- Determines which tools are visible to the agent (included in the tool list sent to the LLM)
permission:
- Controls whether a tool call is allowed, denied, or requires approval
In practice this means that even if a tool is denied via permission, it is still included in the tool definitions sent to the model.
For workflows that rely on strict agent capabilities, this creates two problems:
-
Context bloat
Agents receive many tools they will never use, increasing prompt size. -
Planning noise
The model may still consider tools that will later be blocked by permissions.
Currently the deprecated tools field is still useful because it removes unused tools from the request entirely.
Questions:
- Is there a planned replacement for controlling tool visibility (not just execution permissions)?
- Will
permission: denyeventually hide tools from the LLM, or will all tools always remain exposed? - If
toolsis eventually removed, what is the recommended way to keep agent toolsets minimal?
Example of the problem:
{
"permission": {
"bash": "deny"
}
}The bash tool is still included in the tool list sent to the model (according to what i see in the documentation), even though it cannot execute.
For multi-agent setups (planner, researcher, coder, etc.), controlling which tools are exposed to each agent is important for both security and context efficiency.
Would appreciate clarification on the intended long-term design.
Thanks!
Activity
cebecifaruk commentedon Apr 29, 2026
Any updates?
SkyElianneLavoie commentedon May 28, 2026
I really hope this doesn't get removed from OpenCode!
We use a few MCP servers that have way too many tools, and we only want to give our agent access to a few handful of them.
Setting "deny" on all the tools of the server doesn't fix our issue of the context getting overloaded!
I'm confused why it's even being deprecated...
permissionandtoolsseem to do two distinct things: one declares whether an agent can use a tool, and the other declares whether it knows of the tool.If we still want to deprecate
tools, could we add a new value to thepermissionconfig? Adding a "hidden" keyword that fully removes the tool from the agent's available tools?DasLeo commentedon Jun 10, 2026
We also have models gotten into the wrong direction simply by having tool description in their prompt.
We def. need to keep the
toolsconfig to prevent models from jumping into implementation during the plan phase.rekram1-node commentedon Jun 23, 2026
Automated: closing question issues.
GitHub Issues are for bug reports and feature requests only — questions / support are no longer tracked here.
If this is a bug or feature request, please open a new issue with the appropriate template.