standalone websearch extension (#23823)

## Summary

Add the extension-backed standalone `web.run` tool so Codex can call the
standalone search endpoint through the `codex-api` search client and
return its encrypted output to Responses.

- gate the new tool behind `standalone_web_search`
- install the extension in the app-server thread registry and hide
hosted `web_search` when standalone search is enabled for OpenAI
providers so the two paths stay mutually exclusive
- build search context from persisted history using a small tail
heuristic: previous user message, assistant text between the last two
user turns capped at about 1k tokens, and current user message

## Test Plan

- `cargo test -p codex-web-search-extension`
- `cargo test -p codex-api`
- `cargo test -p codex-core
hosted_tools_follow_provider_auth_model_and_config_gates`

codex-rs/app-server/Cargo.toml

@@ -55,6 +55,7 @@ codex-file-search = { workspace = true }
 codex-chatgpt = { workspace = true }
 codex-login = { workspace = true }
 codex-memories-extension = { workspace = true }
+codex-web-search-extension = { workspace = true }
 codex-memories-write = { workspace = true }
 codex-mcp = { workspace = true }
 codex-model-provider = { workspace = true }

codex-rs/app-server/src/extensions.rs

@@ -12,6 +12,7 @@ use codex_extension_api::AgentSpawner;
 use codex_extension_api::ExtensionEventSink;
 use codex_extension_api::ExtensionRegistry;
 use codex_extension_api::ExtensionRegistryBuilder;
+use codex_login::AuthManager;
 use codex_protocol::ThreadId;
 use codex_protocol::error::CodexErr;
 use codex_protocol::protocol::Event;
@@ -22,6 +23,7 @@ use crate::outgoing_message::OutgoingMessageSender;
 pub(crate) fn thread_extensions<S>(
     guardian_agent_spawner: S,
     event_sink: Arc<dyn ExtensionEventSink>,
+    auth_manager: Arc<AuthManager>,
 ) -> Arc<ExtensionRegistry<Config>>
 where
     S: AgentSpawner<StartThreadOptions, Spawned = NewThread, Error = CodexErr> + 'static,
@@ -29,6 +31,7 @@ where
     let mut builder = ExtensionRegistryBuilder::<Config>::with_event_sink(event_sink);
     codex_guardian::install(&mut builder, guardian_agent_spawner);
     codex_memories_extension::install(&mut builder, codex_otel::global());
+    codex_web_search_extension::install(&mut builder, auth_manager);
     Arc::new(builder.build())
 }
 

codex-rs/app-server/src/mcp_refresh.rs

@@ -184,15 +184,16 @@ mod tests {
         let thread_manager = Arc::new_cyclic(|thread_manager| {
             ThreadManager::new(
                 &good_config,
-                auth_manager,
+                auth_manager.clone(),
                 SessionSource::Exec,
                 Arc::new(EnvironmentManager::default_for_tests()),
                 thread_extensions(
                     guardian_agent_spawner(thread_manager.clone()),
                     Arc::new(NoopExtensionEventSink),
+                    auth_manager.clone(),
                 ),
                 /*analytics_events_client*/ None,
-                thread_store,
+                Arc::clone(&thread_store),
                 Some(state_db.clone()),
                 "11111111-1111-4111-8111-111111111111".to_string(),
                 /*attestation_provider*/ None,

codex-rs/app-server/src/message_processor.rs

@@ -314,6 +314,7 @@ impl MessageProcessor {
                 thread_extensions(
                     guardian_agent_spawner(thread_manager.clone()),
                     app_server_extension_event_sink(outgoing.clone()),
+                    auth_manager.clone(),
                 ),
                 Some(analytics_events_client.clone()),
                 Arc::clone(&thread_store),

codex-rs/app-server/tests/suite/v2/mod.rs

@@ -68,4 +68,5 @@ mod turn_interrupt;
 mod turn_start;
 mod turn_start_zsh_fork;
 mod turn_steer;
+mod web_search;
 mod windows_sandbox_setup;

codex-rs/app-server/tests/suite/v2/web_search.rs

@@ -0,0 +1,216 @@
+use std::path::Path;
+use std::time::Duration;
+
+use anyhow::Context;
+use anyhow::Result;
+use app_test_support::ChatGptAuthFixture;
+use app_test_support::McpProcess;
+use app_test_support::to_response;
+use app_test_support::write_chatgpt_auth;
+use codex_app_server_protocol::JSONRPCResponse;
+use codex_app_server_protocol::RequestId;
+use codex_app_server_protocol::ThreadStartParams;
+use codex_app_server_protocol::ThreadStartResponse;
+use codex_app_server_protocol::TurnStartParams;
+use codex_app_server_protocol::TurnStartResponse;
+use codex_app_server_protocol::UserInput as V2UserInput;
+use codex_config::types::AuthCredentialsStoreMode;
+use core_test_support::responses;
+use pretty_assertions::assert_eq;
+use serde_json::Value;
+use serde_json::json;
+use tempfile::TempDir;
+use tokio::time::timeout;
+use wiremock::Mock;
+use wiremock::MockServer;
+use wiremock::ResponseTemplate;
+use wiremock::matchers::method;
+use wiremock::matchers::path;
+
+// macOS and Windows Bazel CI can spend tens of seconds starting app-server
+// subprocesses or processing test RPCs under load.
+#[cfg(any(target_os = "macos", windows))]
+const DEFAULT_READ_TIMEOUT: Duration = Duration::from_secs(60);
+#[cfg(not(any(target_os = "macos", windows)))]
+const DEFAULT_READ_TIMEOUT: Duration = Duration::from_secs(10);
+
+#[tokio::test]
+async fn standalone_web_search_round_trips_encrypted_output() -> Result<()> {
+    let call_id = "web-run-1";
+    let server = responses::start_mock_server().await;
+    mount_search_response(&server).await;
+
+    let response_mock = responses::mount_sse_sequence(
+        &server,
+        vec![
+            responses::sse(vec![
+                responses::ev_response_created("resp-1"),
+                responses::ev_function_call_with_namespace(
+                    call_id,
+                    "web",
+                    "run",
+                    &json!({
+                        "search_query": [{"q": "standalone web search"}],
+                    })
+                    .to_string(),
+                ),
+                responses::ev_completed("resp-1"),
+            ]),
+            responses::sse(vec![
+                responses::ev_assistant_message("msg-1", "Done"),
+                responses::ev_completed("resp-2"),
+            ]),
+        ],
+    )
+    .await;
+
+    let codex_home = TempDir::new()?;
+    create_config_toml(codex_home.path(), &server.uri())?;
+    write_chatgpt_auth(
+        codex_home.path(),
+        ChatGptAuthFixture::new("access-chatgpt"),
+        AuthCredentialsStoreMode::File,
+    )?;
+
+    let mut mcp = McpProcess::new_with_env(codex_home.path(), &[("OPENAI_API_KEY", None)]).await?;
+    timeout(DEFAULT_READ_TIMEOUT, mcp.initialize()).await??;
+
+    let thread_req = mcp
+        .send_thread_start_request(ThreadStartParams::default())
+        .await?;
+    let thread_resp: JSONRPCResponse = timeout(
+        DEFAULT_READ_TIMEOUT,
+        mcp.read_stream_until_response_message(RequestId::Integer(thread_req)),
+    )
+    .await??;
+    let ThreadStartResponse { thread, .. } = to_response::<ThreadStartResponse>(thread_resp)?;
+
+    let turn_req = mcp
+        .send_turn_start_request(TurnStartParams {
+            thread_id: thread.id,
+            input: vec![V2UserInput::Text {
+                text: "Search the web".to_string(),
+                text_elements: Vec::new(),
+            }],
+            ..Default::default()
+        })
+        .await?;
+    let turn_resp: JSONRPCResponse = timeout(
+        DEFAULT_READ_TIMEOUT,
+        mcp.read_stream_until_response_message(RequestId::Integer(turn_req)),
+    )
+    .await??;
+    let _turn: TurnStartResponse = to_response::<TurnStartResponse>(turn_resp)?;
+
+    timeout(
+        DEFAULT_READ_TIMEOUT,
+        mcp.read_stream_until_notification_message("turn/completed"),
+    )
+    .await??;
+
+    let requests = response_mock.requests();
+    assert_eq!(requests.len(), 2);
+
+    let first_response = requests[0].body_json();
+    assert!(
+        requests[0].tool_by_name("web", "run").is_some(),
+        "web.run should be sent to the model"
+    );
+    assert!(
+        !has_hosted_web_search(&first_response),
+        "standalone web search should replace hosted web search"
+    );
+
+    let search_body = search_request_body(&server).await?;
+    assert_eq!(
+        search_body["commands"],
+        json!({
+            "search_query": [{"q": "standalone web search"}],
+        })
+    );
+    assert_eq!(
+        search_body["input"]
+            .as_array()
+            .context("search input should be an array")?
+            .last(),
+        Some(&json!({
+            "type": "message",
+            "role": "user",
+            "content": [{"type": "input_text", "text": "Search the web"}],
+        }))
+    );
+
+    assert_eq!(
+        requests[1].function_call_output(call_id),
+        json!({
+            "type": "function_call_output",
+            "call_id": call_id,
+            "output": [{
+                "type": "encrypted_content",
+                "encrypted_content": "ciphertext",
+            }],
+        })
+    );
+
+    Ok(())
+}
+
+async fn mount_search_response(server: &MockServer) {
+    Mock::given(method("POST"))
+        .and(path("/api/codex/alpha/search"))
+        .respond_with(ResponseTemplate::new(200).set_body_json(json!({
+            "encrypted_output": "ciphertext",
+        })))
+        .expect(1)
+        .mount(server)
+        .await;
+}
+
+fn has_hosted_web_search(body: &Value) -> bool {
+    body.get("tools")
+        .and_then(Value::as_array)
+        .is_some_and(|tools| {
+            tools
+                .iter()
+                .any(|tool| tool.get("type").and_then(Value::as_str) == Some("web_search"))
+        })
+}
+
+async fn search_request_body(server: &MockServer) -> Result<Value> {
+    server
+        .received_requests()
+        .await
+        .context("failed to fetch received requests")?
+        .into_iter()
+        .find(|request| request.url.path() == "/api/codex/alpha/search")
+        .context("expected standalone search request")?
+        .body_json()
+        .context("search request body should be JSON")
+}
+
+fn create_config_toml(codex_home: &Path, server_uri: &str) -> std::io::Result<()> {
+    std::fs::write(
+        codex_home.join("config.toml"),
+        format!(
+            r#"
+model = "mock-model"
+approval_policy = "never"
+sandbox_mode = "read-only"
+model_provider = "openai-custom"
+chatgpt_base_url = "{server_uri}"
+
+[features]
+standalone_web_search = true
+
+[model_providers.openai-custom]
+name = "OpenAI"
+base_url = "{server_uri}/api/codex"
+wire_api = "responses"
+request_max_retries = 0
+stream_max_retries = 0
+supports_websockets = false
+requires_openai_auth = true
+"#
+        ),
+    )
+}