fix: separate codex mcp into codex mcp-server and codex app-server (#4471)

This is a very large PR with some non-backwards-compatible changes.

Historically, `codex mcp` (or `codex mcp serve`) started a JSON-RPC-ish
server that had two overlapping responsibilities:

- Running an MCP server, providing some basic tool calls.
- Running the app server used to power experiences such as the VS Code
extension.

This PR aims to separate these into distinct concepts:

- `codex mcp-server` for the MCP server
- `codex app-server` for the "application server"

Note `codex mcp` still exists because it already has its own subcommands
for MCP management (`list`, `add`, etc.)

The MCP logic continues to live in `codex-rs/mcp-server` whereas the
refactored app server logic is in the new `codex-rs/app-server` folder.
Note that most of the existing integration tests in
`codex-rs/mcp-server/tests/suite` were actually for the app server, so
all the tests have been moved with the exception of
`codex-rs/mcp-server/tests/suite/mod.rs`.

Because this is already a large diff, I tried not to change more than I
had to, so `codex-rs/app-server/tests/common/mcp_process.rs` still uses
the name `McpProcess` for now, but I will do some mechanical renamings
to things like `AppServer` in subsequent PRs.

While `mcp-server` and `app-server` share some overlapping functionality
(like reading streams of JSONL and dispatching based on message types)
and some differences (completely different message types), I ended up
doing a bit of copypasta between the two crates, as both have somewhat
similar `message_processor.rs` and `outgoing_message.rs` files for now,
though I expect them to diverge more in the near future.

One material change is that of the initialize handshake for `codex
app-server`, as we no longer use the MCP types for that handshake.
Instead, we update `codex-rs/protocol/src/mcp_protocol.rs` to add an
`Initialize` variant to `ClientRequest`, which takes the `ClientInfo`
object we need to update the `USER_AGENT_SUFFIX` in
`codex-rs/app-server/src/message_processor.rs`.

One other material change is in
`codex-rs/app-server/src/codex_message_processor.rs` where I eliminated
a use of the `send_event_as_notification()` method I am generally trying
to deprecate (because it blindly maps an `EventMsg` into a
`JSONNotification`) in favor of `send_server_notification()`, which
takes a `ServerNotification`, as that is intended to be a custom enum of
all notification types supported by the app server. So to make this
update, I had to introduce a new variant of `ServerNotification`,
`SessionConfigured`, which is a non-backwards compatible change with the
old `codex mcp`, and clients will have to be updated after the next
release that contains this PR. Note that
`codex-rs/app-server/tests/suite/list_resume.rs` also had to be update
to reflect this change.

I introduced `codex-rs/utils/json-to-toml/src/lib.rs` as a small utility
crate to avoid some of the copying between `mcp-server` and
`app-server`.

codex-rs/app-server/tests/common/Cargo.toml

@@ -0,0 +1,22 @@
+[package]
+edition = "2024"
+name = "app_test_support"
+version = { workspace = true }
+
+[lib]
+path = "lib.rs"
+
+[dependencies]
+anyhow = { workspace = true }
+assert_cmd = { workspace = true }
+codex-protocol = { workspace = true }
+mcp-types = { workspace = true }
+serde = { workspace = true }
+serde_json = { workspace = true }
+tokio = { workspace = true, features = [
+    "io-std",
+    "macros",
+    "process",
+    "rt-multi-thread",
+] }
+wiremock = { workspace = true }

codex-rs/app-server/tests/common/lib.rs

@@ -0,0 +1,17 @@
+mod mcp_process;
+mod mock_model_server;
+mod responses;
+
+pub use mcp_process::McpProcess;
+use mcp_types::JSONRPCResponse;
+pub use mock_model_server::create_mock_chat_completions_server;
+pub use responses::create_apply_patch_sse_response;
+pub use responses::create_final_assistant_message_sse_response;
+pub use responses::create_shell_sse_response;
+use serde::de::DeserializeOwned;
+
+pub fn to_response<T: DeserializeOwned>(response: JSONRPCResponse) -> anyhow::Result<T> {
+    let value = serde_json::to_value(response.result)?;
+    let codex_response = serde_json::from_value(value)?;
+    Ok(codex_response)
+}

codex-rs/app-server/tests/common/mcp_process.rs

@@ -0,0 +1,477 @@
+use std::path::Path;
+use std::process::Stdio;
+use std::sync::atomic::AtomicI64;
+use std::sync::atomic::Ordering;
+use tokio::io::AsyncBufReadExt;
+use tokio::io::AsyncWriteExt;
+use tokio::io::BufReader;
+use tokio::process::Child;
+use tokio::process::ChildStdin;
+use tokio::process::ChildStdout;
+
+use anyhow::Context;
+use assert_cmd::prelude::*;
+use codex_protocol::mcp_protocol::AddConversationListenerParams;
+use codex_protocol::mcp_protocol::ArchiveConversationParams;
+use codex_protocol::mcp_protocol::CancelLoginChatGptParams;
+use codex_protocol::mcp_protocol::ClientInfo;
+use codex_protocol::mcp_protocol::ClientNotification;
+use codex_protocol::mcp_protocol::GetAuthStatusParams;
+use codex_protocol::mcp_protocol::InitializeParams;
+use codex_protocol::mcp_protocol::InterruptConversationParams;
+use codex_protocol::mcp_protocol::ListConversationsParams;
+use codex_protocol::mcp_protocol::LoginApiKeyParams;
+use codex_protocol::mcp_protocol::NewConversationParams;
+use codex_protocol::mcp_protocol::RemoveConversationListenerParams;
+use codex_protocol::mcp_protocol::ResumeConversationParams;
+use codex_protocol::mcp_protocol::SendUserMessageParams;
+use codex_protocol::mcp_protocol::SendUserTurnParams;
+use codex_protocol::mcp_protocol::SetDefaultModelParams;
+
+use mcp_types::JSONRPC_VERSION;
+use mcp_types::JSONRPCMessage;
+use mcp_types::JSONRPCNotification;
+use mcp_types::JSONRPCRequest;
+use mcp_types::JSONRPCResponse;
+use mcp_types::RequestId;
+use std::process::Command as StdCommand;
+use tokio::process::Command;
+
+pub struct McpProcess {
+    next_request_id: AtomicI64,
+    /// Retain this child process until the client is dropped. The Tokio runtime
+    /// will make a "best effort" to reap the process after it exits, but it is
+    /// not a guarantee. See the `kill_on_drop` documentation for details.
+    #[allow(dead_code)]
+    process: Child,
+    stdin: ChildStdin,
+    stdout: BufReader<ChildStdout>,
+}
+
+impl McpProcess {
+    pub async fn new(codex_home: &Path) -> anyhow::Result<Self> {
+        Self::new_with_env(codex_home, &[]).await
+    }
+
+    /// Creates a new MCP process, allowing tests to override or remove
+    /// specific environment variables for the child process only.
+    ///
+    /// Pass a tuple of (key, Some(value)) to set/override, or (key, None) to
+    /// remove a variable from the child's environment.
+    pub async fn new_with_env(
+        codex_home: &Path,
+        env_overrides: &[(&str, Option<&str>)],
+    ) -> anyhow::Result<Self> {
+        // Use assert_cmd to locate the binary path and then switch to tokio::process::Command
+        let std_cmd = StdCommand::cargo_bin("codex-app-server")
+            .context("should find binary for codex-mcp-server")?;
+
+        let program = std_cmd.get_program().to_owned();
+
+        let mut cmd = Command::new(program);
+
+        cmd.stdin(Stdio::piped());
+        cmd.stdout(Stdio::piped());
+        cmd.stderr(Stdio::piped());
+        cmd.env("CODEX_HOME", codex_home);
+        cmd.env("RUST_LOG", "debug");
+
+        for (k, v) in env_overrides {
+            match v {
+                Some(val) => {
+                    cmd.env(k, val);
+                }
+                None => {
+                    cmd.env_remove(k);
+                }
+            }
+        }
+
+        let mut process = cmd
+            .kill_on_drop(true)
+            .spawn()
+            .context("codex-mcp-server proc should start")?;
+        let stdin = process
+            .stdin
+            .take()
+            .ok_or_else(|| anyhow::format_err!("mcp should have stdin fd"))?;
+        let stdout = process
+            .stdout
+            .take()
+            .ok_or_else(|| anyhow::format_err!("mcp should have stdout fd"))?;
+        let stdout = BufReader::new(stdout);
+
+        // Forward child's stderr to our stderr so failures are visible even
+        // when stdout/stderr are captured by the test harness.
+        if let Some(stderr) = process.stderr.take() {
+            let mut stderr_reader = BufReader::new(stderr).lines();
+            tokio::spawn(async move {
+                while let Ok(Some(line)) = stderr_reader.next_line().await {
+                    eprintln!("[mcp stderr] {line}");
+                }
+            });
+        }
+        Ok(Self {
+            next_request_id: AtomicI64::new(0),
+            process,
+            stdin,
+            stdout,
+        })
+    }
+
+    /// Performs the initialization handshake with the MCP server.
+    pub async fn initialize(&mut self) -> anyhow::Result<()> {
+        let params = Some(serde_json::to_value(InitializeParams {
+            client_info: ClientInfo {
+                name: "codex-app-server-tests".to_string(),
+                title: None,
+                version: "0.1.0".to_string(),
+            },
+        })?);
+        let req_id = self.send_request("initialize", params).await?;
+        let initialized = self.read_jsonrpc_message().await?;
+        let JSONRPCMessage::Response(response) = initialized else {
+            unreachable!("expected JSONRPCMessage::Response for initialize, got {initialized:?}");
+        };
+        if response.id != RequestId::Integer(req_id) {
+            anyhow::bail!(
+                "initialize response id mismatch: expected {}, got {:?}",
+                req_id,
+                response.id
+            );
+        }
+
+        // Send notifications/initialized to ack the response.
+        self.send_notification(ClientNotification::Initialized)
+            .await?;
+
+        Ok(())
+    }
+
+    /// Send a `newConversation` JSON-RPC request.
+    pub async fn send_new_conversation_request(
+        &mut self,
+        params: NewConversationParams,
+    ) -> anyhow::Result<i64> {
+        let params = Some(serde_json::to_value(params)?);
+        self.send_request("newConversation", params).await
+    }
+
+    /// Send an `archiveConversation` JSON-RPC request.
+    pub async fn send_archive_conversation_request(
+        &mut self,
+        params: ArchiveConversationParams,
+    ) -> anyhow::Result<i64> {
+        let params = Some(serde_json::to_value(params)?);
+        self.send_request("archiveConversation", params).await
+    }
+
+    /// Send an `addConversationListener` JSON-RPC request.
+    pub async fn send_add_conversation_listener_request(
+        &mut self,
+        params: AddConversationListenerParams,
+    ) -> anyhow::Result<i64> {
+        let params = Some(serde_json::to_value(params)?);
+        self.send_request("addConversationListener", params).await
+    }
+
+    /// Send a `sendUserMessage` JSON-RPC request with a single text item.
+    pub async fn send_send_user_message_request(
+        &mut self,
+        params: SendUserMessageParams,
+    ) -> anyhow::Result<i64> {
+        // Wire format expects variants in camelCase; text item uses external tagging.
+        let params = Some(serde_json::to_value(params)?);
+        self.send_request("sendUserMessage", params).await
+    }
+
+    /// Send a `removeConversationListener` JSON-RPC request.
+    pub async fn send_remove_conversation_listener_request(
+        &mut self,
+        params: RemoveConversationListenerParams,
+    ) -> anyhow::Result<i64> {
+        let params = Some(serde_json::to_value(params)?);
+        self.send_request("removeConversationListener", params)
+            .await
+    }
+
+    /// Send a `sendUserTurn` JSON-RPC request.
+    pub async fn send_send_user_turn_request(
+        &mut self,
+        params: SendUserTurnParams,
+    ) -> anyhow::Result<i64> {
+        let params = Some(serde_json::to_value(params)?);
+        self.send_request("sendUserTurn", params).await
+    }
+
+    /// Send a `interruptConversation` JSON-RPC request.
+    pub async fn send_interrupt_conversation_request(
+        &mut self,
+        params: InterruptConversationParams,
+    ) -> anyhow::Result<i64> {
+        let params = Some(serde_json::to_value(params)?);
+        self.send_request("interruptConversation", params).await
+    }
+
+    /// Send a `getAuthStatus` JSON-RPC request.
+    pub async fn send_get_auth_status_request(
+        &mut self,
+        params: GetAuthStatusParams,
+    ) -> anyhow::Result<i64> {
+        let params = Some(serde_json::to_value(params)?);
+        self.send_request("getAuthStatus", params).await
+    }
+
+    /// Send a `getUserSavedConfig` JSON-RPC request.
+    pub async fn send_get_user_saved_config_request(&mut self) -> anyhow::Result<i64> {
+        self.send_request("getUserSavedConfig", None).await
+    }
+
+    /// Send a `getUserAgent` JSON-RPC request.
+    pub async fn send_get_user_agent_request(&mut self) -> anyhow::Result<i64> {
+        self.send_request("getUserAgent", None).await
+    }
+
+    /// Send a `userInfo` JSON-RPC request.
+    pub async fn send_user_info_request(&mut self) -> anyhow::Result<i64> {
+        self.send_request("userInfo", None).await
+    }
+
+    /// Send a `setDefaultModel` JSON-RPC request.
+    pub async fn send_set_default_model_request(
+        &mut self,
+        params: SetDefaultModelParams,
+    ) -> anyhow::Result<i64> {
+        let params = Some(serde_json::to_value(params)?);
+        self.send_request("setDefaultModel", params).await
+    }
+
+    /// Send a `listConversations` JSON-RPC request.
+    pub async fn send_list_conversations_request(
+        &mut self,
+        params: ListConversationsParams,
+    ) -> anyhow::Result<i64> {
+        let params = Some(serde_json::to_value(params)?);
+        self.send_request("listConversations", params).await
+    }
+
+    /// Send a `resumeConversation` JSON-RPC request.
+    pub async fn send_resume_conversation_request(
+        &mut self,
+        params: ResumeConversationParams,
+    ) -> anyhow::Result<i64> {
+        let params = Some(serde_json::to_value(params)?);
+        self.send_request("resumeConversation", params).await
+    }
+
+    /// Send a `loginApiKey` JSON-RPC request.
+    pub async fn send_login_api_key_request(
+        &mut self,
+        params: LoginApiKeyParams,
+    ) -> anyhow::Result<i64> {
+        let params = Some(serde_json::to_value(params)?);
+        self.send_request("loginApiKey", params).await
+    }
+
+    /// Send a `loginChatGpt` JSON-RPC request.
+    pub async fn send_login_chat_gpt_request(&mut self) -> anyhow::Result<i64> {
+        self.send_request("loginChatGpt", None).await
+    }
+
+    /// Send a `cancelLoginChatGpt` JSON-RPC request.
+    pub async fn send_cancel_login_chat_gpt_request(
+        &mut self,
+        params: CancelLoginChatGptParams,
+    ) -> anyhow::Result<i64> {
+        let params = Some(serde_json::to_value(params)?);
+        self.send_request("cancelLoginChatGpt", params).await
+    }
+
+    /// Send a `logoutChatGpt` JSON-RPC request.
+    pub async fn send_logout_chat_gpt_request(&mut self) -> anyhow::Result<i64> {
+        self.send_request("logoutChatGpt", None).await
+    }
+
+    /// Send a `fuzzyFileSearch` JSON-RPC request.
+    pub async fn send_fuzzy_file_search_request(
+        &mut self,
+        query: &str,
+        roots: Vec<String>,
+        cancellation_token: Option<String>,
+    ) -> anyhow::Result<i64> {
+        let mut params = serde_json::json!({
+            "query": query,
+            "roots": roots,
+        });
+        if let Some(token) = cancellation_token {
+            params["cancellationToken"] = serde_json::json!(token);
+        }
+        self.send_request("fuzzyFileSearch", Some(params)).await
+    }
+
+    async fn send_request(
+        &mut self,
+        method: &str,
+        params: Option<serde_json::Value>,
+    ) -> anyhow::Result<i64> {
+        let request_id = self.next_request_id.fetch_add(1, Ordering::Relaxed);
+
+        let message = JSONRPCMessage::Request(JSONRPCRequest {
+            jsonrpc: JSONRPC_VERSION.into(),
+            id: RequestId::Integer(request_id),
+            method: method.to_string(),
+            params,
+        });
+        self.send_jsonrpc_message(message).await?;
+        Ok(request_id)
+    }
+
+    pub async fn send_response(
+        &mut self,
+        id: RequestId,
+        result: serde_json::Value,
+    ) -> anyhow::Result<()> {
+        self.send_jsonrpc_message(JSONRPCMessage::Response(JSONRPCResponse {
+            jsonrpc: JSONRPC_VERSION.into(),
+            id,
+            result,
+        }))
+        .await
+    }
+
+    pub async fn send_notification(
+        &mut self,
+        notification: ClientNotification,
+    ) -> anyhow::Result<()> {
+        let value = serde_json::to_value(notification)?;
+        self.send_jsonrpc_message(JSONRPCMessage::Notification(JSONRPCNotification {
+            jsonrpc: JSONRPC_VERSION.into(),
+            method: value
+                .get("method")
+                .and_then(|m| m.as_str())
+                .ok_or_else(|| anyhow::format_err!("notification missing method field"))?
+                .to_string(),
+            params: value.get("params").cloned(),
+        }))
+        .await
+    }
+
+    async fn send_jsonrpc_message(&mut self, message: JSONRPCMessage) -> anyhow::Result<()> {
+        eprintln!("writing message to stdin: {message:?}");
+        let payload = serde_json::to_string(&message)?;
+        self.stdin.write_all(payload.as_bytes()).await?;
+        self.stdin.write_all(b"\n").await?;
+        self.stdin.flush().await?;
+        Ok(())
+    }
+
+    async fn read_jsonrpc_message(&mut self) -> anyhow::Result<JSONRPCMessage> {
+        let mut line = String::new();
+        self.stdout.read_line(&mut line).await?;
+        let message = serde_json::from_str::<JSONRPCMessage>(&line)?;
+        eprintln!("read message from stdout: {message:?}");
+        Ok(message)
+    }
+
+    pub async fn read_stream_until_request_message(&mut self) -> anyhow::Result<JSONRPCRequest> {
+        eprintln!("in read_stream_until_request_message()");
+
+        loop {
+            let message = self.read_jsonrpc_message().await?;
+
+            match message {
+                JSONRPCMessage::Notification(_) => {
+                    eprintln!("notification: {message:?}");
+                }
+                JSONRPCMessage::Request(jsonrpc_request) => {
+                    return Ok(jsonrpc_request);
+                }
+                JSONRPCMessage::Error(_) => {
+                    anyhow::bail!("unexpected JSONRPCMessage::Error: {message:?}");
+                }
+                JSONRPCMessage::Response(_) => {
+                    anyhow::bail!("unexpected JSONRPCMessage::Response: {message:?}");
+                }
+            }
+        }
+    }
+
+    pub async fn read_stream_until_response_message(
+        &mut self,
+        request_id: RequestId,
+    ) -> anyhow::Result<JSONRPCResponse> {
+        eprintln!("in read_stream_until_response_message({request_id:?})");
+
+        loop {
+            let message = self.read_jsonrpc_message().await?;
+            match message {
+                JSONRPCMessage::Notification(_) => {
+                    eprintln!("notification: {message:?}");
+                }
+                JSONRPCMessage::Request(_) => {
+                    anyhow::bail!("unexpected JSONRPCMessage::Request: {message:?}");
+                }
+                JSONRPCMessage::Error(_) => {
+                    anyhow::bail!("unexpected JSONRPCMessage::Error: {message:?}");
+                }
+                JSONRPCMessage::Response(jsonrpc_response) => {
+                    if jsonrpc_response.id == request_id {
+                        return Ok(jsonrpc_response);
+                    }
+                }
+            }
+        }
+    }
+
+    pub async fn read_stream_until_error_message(
+        &mut self,
+        request_id: RequestId,
+    ) -> anyhow::Result<mcp_types::JSONRPCError> {
+        loop {
+            let message = self.read_jsonrpc_message().await?;
+            match message {
+                JSONRPCMessage::Notification(_) => {
+                    eprintln!("notification: {message:?}");
+                }
+                JSONRPCMessage::Request(_) => {
+                    anyhow::bail!("unexpected JSONRPCMessage::Request: {message:?}");
+                }
+                JSONRPCMessage::Response(_) => {
+                    // Keep scanning; we're waiting for an error with matching id.
+                }
+                JSONRPCMessage::Error(err) => {
+                    if err.id == request_id {
+                        return Ok(err);
+                    }
+                }
+            }
+        }
+    }
+
+    pub async fn read_stream_until_notification_message(
+        &mut self,
+        method: &str,
+    ) -> anyhow::Result<JSONRPCNotification> {
+        eprintln!("in read_stream_until_notification_message({method})");
+
+        loop {
+            let message = self.read_jsonrpc_message().await?;
+            match message {
+                JSONRPCMessage::Notification(notification) => {
+                    if notification.method == method {
+                        return Ok(notification);
+                    }
+                }
+                JSONRPCMessage::Request(_) => {
+                    anyhow::bail!("unexpected JSONRPCMessage::Request: {message:?}");
+                }
+                JSONRPCMessage::Error(_) => {
+                    anyhow::bail!("unexpected JSONRPCMessage::Error: {message:?}");
+                }
+                JSONRPCMessage::Response(_) => {
+                    anyhow::bail!("unexpected JSONRPCMessage::Response: {message:?}");
+                }
+            }
+        }
+    }
+}

codex-rs/app-server/tests/common/mock_model_server.rs

@@ -0,0 +1,47 @@
+use std::sync::atomic::AtomicUsize;
+use std::sync::atomic::Ordering;
+
+use wiremock::Mock;
+use wiremock::MockServer;
+use wiremock::Respond;
+use wiremock::ResponseTemplate;
+use wiremock::matchers::method;
+use wiremock::matchers::path;
+
+/// Create a mock server that will provide the responses, in order, for
+/// requests to the `/v1/chat/completions` endpoint.
+pub async fn create_mock_chat_completions_server(responses: Vec<String>) -> MockServer {
+    let server = MockServer::start().await;
+
+    let num_calls = responses.len();
+    let seq_responder = SeqResponder {
+        num_calls: AtomicUsize::new(0),
+        responses,
+    };
+
+    Mock::given(method("POST"))
+        .and(path("/v1/chat/completions"))
+        .respond_with(seq_responder)
+        .expect(num_calls as u64)
+        .mount(&server)
+        .await;
+
+    server
+}
+
+struct SeqResponder {
+    num_calls: AtomicUsize,
+    responses: Vec<String>,
+}
+
+impl Respond for SeqResponder {
+    fn respond(&self, _: &wiremock::Request) -> ResponseTemplate {
+        let call_num = self.num_calls.fetch_add(1, Ordering::SeqCst);
+        match self.responses.get(call_num) {
+            Some(response) => ResponseTemplate::new(200)
+                .insert_header("content-type", "text/event-stream")
+                .set_body_raw(response.clone(), "text/event-stream"),
+            None => panic!("no response for {call_num}"),
+        }
+    }
+}

codex-rs/app-server/tests/common/responses.rs

@@ -0,0 +1,95 @@
+use serde_json::json;
+use std::path::Path;
+
+pub fn create_shell_sse_response(
+    command: Vec<String>,
+    workdir: Option<&Path>,
+    timeout_ms: Option<u64>,
+    call_id: &str,
+) -> anyhow::Result<String> {
+    // The `arguments`` for the `shell` tool is a serialized JSON object.
+    let tool_call_arguments = serde_json::to_string(&json!({
+        "command": command,
+        "workdir": workdir.map(|w| w.to_string_lossy()),
+        "timeout": timeout_ms
+    }))?;
+    let tool_call = json!({
+        "choices": [
+            {
+                "delta": {
+                    "tool_calls": [
+                        {
+                            "id": call_id,
+                            "function": {
+                                "name": "shell",
+                                "arguments": tool_call_arguments
+                            }
+                        }
+                    ]
+                },
+                "finish_reason": "tool_calls"
+            }
+        ]
+    });
+
+    let sse = format!(
+        "data: {}\n\ndata: DONE\n\n",
+        serde_json::to_string(&tool_call)?
+    );
+    Ok(sse)
+}
+
+pub fn create_final_assistant_message_sse_response(message: &str) -> anyhow::Result<String> {
+    let assistant_message = json!({
+        "choices": [
+            {
+                "delta": {
+                    "content": message
+                },
+                "finish_reason": "stop"
+            }
+        ]
+    });
+
+    let sse = format!(
+        "data: {}\n\ndata: DONE\n\n",
+        serde_json::to_string(&assistant_message)?
+    );
+    Ok(sse)
+}
+
+pub fn create_apply_patch_sse_response(
+    patch_content: &str,
+    call_id: &str,
+) -> anyhow::Result<String> {
+    // Use shell command to call apply_patch with heredoc format
+    let shell_command = format!("apply_patch <<'EOF'\n{patch_content}\nEOF");
+    let tool_call_arguments = serde_json::to_string(&json!({
+        "command": ["bash", "-lc", shell_command]
+    }))?;
+
+    let tool_call = json!({
+        "choices": [
+            {
+                "delta": {
+                    "tool_calls": [
+                        {
+                            "id": call_id,
+                            "function": {
+                                "name": "shell",
+                                "arguments": tool_call_arguments
+                            }
+                        }
+                    ]
+                },
+                "finish_reason": "tool_calls"
+            }
+        ]
+    });
+
+    let sse = format!(
+        "data: {}\n\ndata: DONE\n\n",
+        serde_json::to_string(&tool_call)?
+    );
+    Ok(sse)
+}