Files
codex/codex-rs/core/src/tools/handlers/plan.rs
T
Gabriel Peal b0bdc04c30 [MCP] Render MCP tool call result images to the model (#5600)
It's pretty amazing we have gotten here without the ability for the
model to see image content from MCP tool calls.

This PR builds off of 4391 and fixes #4819. I would like @KKcorps to get
adequete credit here but I also want to get this fix in ASAP so I gave
him a week to update it and haven't gotten a response so I'm going to
take it across the finish line.


This test highlights how absured the current situation is. I asked the
model to read this image using the Chrome MCP
<img width="2378" height="674" alt="image"
src="https://github.com/user-attachments/assets/9ef52608-72a2-4423-9f5e-7ae36b2b56e0"
/>

After this change, it correctly outputs:
> Captured the page: image dhows a dark terminal-style UI labeled
`OpenAI Codex (v0.0.0)` with prompt `model: gpt-5-codex medium` and
working directory `/codex/codex-rs`
(and more)  

Before this change, it said:
> Took the full-page screenshot you asked for. It shows a long,
horizontally repeating pattern of stylized people in orange, light-blue,
and mustard clothing, holding hands in alternating poses against a white
background. No text or other graphics-just rows of flat illustration
stretching off to the right.

Without this change, the Figma, Playwright, Chrome, and other visual MCP
servers are pretty much entirely useless.

I tested this change with the openai respones api as well as a third
party completions api
2025-10-27 17:55:57 -04:00

118 lines
3.9 KiB
Rust

use crate::client_common::tools::ResponsesApiTool;
use crate::client_common::tools::ToolSpec;
use crate::codex::Session;
use crate::codex::TurnContext;
use crate::function_tool::FunctionCallError;
use crate::tools::context::ToolInvocation;
use crate::tools::context::ToolOutput;
use crate::tools::context::ToolPayload;
use crate::tools::registry::ToolHandler;
use crate::tools::registry::ToolKind;
use crate::tools::spec::JsonSchema;
use async_trait::async_trait;
use codex_protocol::plan_tool::UpdatePlanArgs;
use codex_protocol::protocol::EventMsg;
use std::collections::BTreeMap;
use std::sync::LazyLock;
pub struct PlanHandler;
pub static PLAN_TOOL: LazyLock<ToolSpec> = LazyLock::new(|| {
let mut plan_item_props = BTreeMap::new();
plan_item_props.insert("step".to_string(), JsonSchema::String { description: None });
plan_item_props.insert(
"status".to_string(),
JsonSchema::String {
description: Some("One of: pending, in_progress, completed".to_string()),
},
);
let plan_items_schema = JsonSchema::Array {
description: Some("The list of steps".to_string()),
items: Box::new(JsonSchema::Object {
properties: plan_item_props,
required: Some(vec!["step".to_string(), "status".to_string()]),
additional_properties: Some(false.into()),
}),
};
let mut properties = BTreeMap::new();
properties.insert(
"explanation".to_string(),
JsonSchema::String { description: None },
);
properties.insert("plan".to_string(), plan_items_schema);
ToolSpec::Function(ResponsesApiTool {
name: "update_plan".to_string(),
description: r#"Updates the task plan.
Provide an optional explanation and a list of plan items, each with a step and status.
At most one step can be in_progress at a time.
"#
.to_string(),
strict: false,
parameters: JsonSchema::Object {
properties,
required: Some(vec!["plan".to_string()]),
additional_properties: Some(false.into()),
},
})
});
#[async_trait]
impl ToolHandler for PlanHandler {
fn kind(&self) -> ToolKind {
ToolKind::Function
}
async fn handle(&self, invocation: ToolInvocation) -> Result<ToolOutput, FunctionCallError> {
let ToolInvocation {
session,
turn,
call_id,
payload,
..
} = invocation;
let arguments = match payload {
ToolPayload::Function { arguments } => arguments,
_ => {
return Err(FunctionCallError::RespondToModel(
"update_plan handler received unsupported payload".to_string(),
));
}
};
let content =
handle_update_plan(session.as_ref(), turn.as_ref(), arguments, call_id).await?;
Ok(ToolOutput::Function {
content,
content_items: None,
success: Some(true),
})
}
}
/// This function doesn't do anything useful. However, it gives the model a structured way to record its plan that clients can read and render.
/// So it's the _inputs_ to this function that are useful to clients, not the outputs and neither are actually useful for the model other
/// than forcing it to come up and document a plan (TBD how that affects performance).
pub(crate) async fn handle_update_plan(
session: &Session,
turn_context: &TurnContext,
arguments: String,
_call_id: String,
) -> Result<String, FunctionCallError> {
let args = parse_update_plan_arguments(&arguments)?;
session
.send_event(turn_context, EventMsg::PlanUpdate(args))
.await;
Ok("Plan updated".to_string())
}
fn parse_update_plan_arguments(arguments: &str) -> Result<UpdatePlanArgs, FunctionCallError> {
serde_json::from_str::<UpdatePlanArgs>(arguments).map_err(|e| {
FunctionCallError::RespondToModel(format!("failed to parse function arguments: {e}"))
})
}