From fcc4ca552f7753c537fe4b8af4276caf8edff762 Mon Sep 17 00:00:00 2001
From: jif <jif@openai.com>
Date: Wed, 5 Aug 2026 09:33:38 +0000
Subject: [PATCH] Preserve longer MCP source descriptions in tool search
 (#37066)

## Why

The 4 KiB aggregate limit can truncate source metadata before it reaches the
model through the tool search description.

## What changed

Raise the aggregate source-description limit to 512 KiB while retaining the
existing character-boundary-safe truncation and bounded output.

## Testing

Expand unit and RMCP round-trip coverage to verify that substantially longer
Unicode descriptions remain complete within the new limit.

GitOrigin-RevId: 002760cdd2e8741e531b45f75f8624ef84334d24
---
 codex-rs/core/src/tools/handlers/tool_search_spec.rs | 6 ++++--
 codex-rs/core/tests/suite/rmcp_client.rs             | 4 ++--
 2 files changed, 6 insertions(+), 4 deletions(-)

diff --git a/codex-rs/core/src/tools/handlers/tool_search_spec.rs b/codex-rs/core/src/tools/handlers/tool_search_spec.rs
index cdcf9ff823..c607ae0762 100644
--- a/codex-rs/core/src/tools/handlers/tool_search_spec.rs
+++ b/codex-rs/core/src/tools/handlers/tool_search_spec.rs
@@ -5,7 +5,7 @@ use codex_tools::ToolSpec;
 use codex_utils_string::take_bytes_at_char_boundary;
 use std::collections::BTreeMap;
 
-const MAX_TOOL_SEARCH_SOURCE_DESCRIPTION_BYTES: usize = 4 * 1024;
+const MAX_TOOL_SEARCH_SOURCE_DESCRIPTION_BYTES: usize = 512 * 1024;
 
 #[derive(Clone, Copy, Debug, PartialEq, Eq)]
 pub(crate) enum ToolSearchSourceListing {
@@ -176,10 +176,11 @@ mod tests {
 
     #[test]
     fn create_tool_search_tool_bounds_aggregate_source_descriptions() {
+        let long_description = "🦀".repeat(20_000);
         let sources = (0..8)
             .map(|index| ToolSearchSourceInfo {
                 name: format!("source-{index:02}"),
-                description: Some("🦀".repeat(300)),
+                description: Some(long_description.clone()),
             })
             .collect::<Vec<_>>();
         let ToolSpec::ToolSearch { description, .. } = create_tool_search_tool(
@@ -198,6 +199,7 @@ mod tests {
             .expect("tool search should retain its discovery instructions");
         assert!(source_descriptions.len() <= MAX_TOOL_SEARCH_SOURCE_DESCRIPTION_BYTES);
         assert!(source_descriptions.starts_with("- source-00: 🦀"));
+        assert!(source_descriptions.contains(&long_description));
         let advertised_names = source_descriptions
             .lines()
             .map(|line| {
diff --git a/codex-rs/core/tests/suite/rmcp_client.rs b/codex-rs/core/tests/suite/rmcp_client.rs
index 4c6d2032d4..6bb23a3221 100644
--- a/codex-rs/core/tests/suite/rmcp_client.rs
+++ b/codex-rs/core/tests/suite/rmcp_client.rs
@@ -620,7 +620,7 @@ async fn stdio_server_round_trip() -> anyhow::Result<()> {
     .await;
 
     let expected_env_value = "propagated-env";
-    let expected_description = format!("{}🦀keep the complete MCP metadata", "é".repeat(499));
+    let expected_description = format!("{}🦀keep the complete MCP metadata", "é".repeat(11_000));
     let instructions = expected_description.clone();
     let rmcp_test_server_bin = remote_aware_stdio_server_bin()?;
 
@@ -713,7 +713,7 @@ async fn stdio_server_round_trip() -> anyhow::Result<()> {
         .and_then(|tool| tool.get("description").and_then(Value::as_str))
         .expect("the model should receive a tool search description");
     assert!(
-        search_description.len() < 5 * 1024,
+        search_description.len() < 513 * 1024,
         "the complete tool search description must remain bounded"
     );
     assert!(search_description.contains(&format!("- rmcp: {expected_description}")));