Remove cumulative MCP attribution from requests and rollouts (#62589)

## What changed

Stop collecting cumulative MCP tool attribution, sending it in Responses
API request metadata, and writing it to rollout checkpoints.

Use `model_search_state` for search provider history in new rollouts. When
reading older rollouts, decode `mcp_attribution` only to determine whether
Search MCP may have been used. Treat malformed or error attribution as
possible Search MCP use so native search fallback stays disabled.

## Testing

Add a history test that distinguishes legacy Search MCP use, an explicit
record of no use, and missing attribution.

GitOrigin-RevId: a145b5b6b7d2f423d989af10623ef21120f49d4d
This commit is contained in:
Owen Lin
2026-11-19 21:06:08 +00:00
committed by copyberry
parent 5f8c36e836
commit 30c69e60a3
24 changed files with 98 additions and 1248 deletions
-75
View File
@@ -5,7 +5,6 @@
//! `schemars`) so they can be embedded in Codex's own protocol structures.
use schemars::JsonSchema;
use serde::Deserialize;
use serde::Deserializer;
use serde::Serialize;
use std::collections::HashMap;
use ts_rs::TS;
@@ -64,80 +63,6 @@ pub fn is_node_repl_backed_tool(name: &str, namespace: Option<&str>) -> bool {
.is_some_and(|(server, _)| is_node_repl_backed_server(server))
}
/// Producer-reported completeness of the thread's MCP tools/call attribution.
#[derive(Clone, Copy, Debug, Default, PartialEq, Eq, Serialize, Deserialize, JsonSchema)]
#[serde(rename_all = "snake_case")]
pub enum McpAttributionStatus {
#[default]
None,
Complete,
AttributionError,
}
/// Bounded diagnostics for an attribution error, never a policy decision.
///
/// Recorder reasons identify the first observed error and survive checkpoints.
/// Payload reasons describe a request-serialization fallback and are not checkpointed.
#[derive(Clone, Copy, Debug, PartialEq, Eq, Serialize, Deserialize, JsonSchema)]
#[serde(rename_all = "snake_case")]
#[non_exhaustive]
pub enum McpAttributionErrorReason {
HistoryMissingCheckpoint,
CheckpointInvalid,
CheckpointSourceConflict,
SourceInvalid,
RecorderPoisoned,
/// An earlier error checkpoint did not retain its cause.
RestoredErrorUnknown,
PayloadTooLarge,
SerializationFailed,
#[serde(other)]
Unknown,
}
/// One source and the first runtime turn in which Codex recorded it.
///
/// Server and tool names are retained when no stable identifier is available;
/// they are not guaranteed to be globally unique or independently authenticated.
#[derive(Clone, Debug, PartialEq, Eq, Serialize, Deserialize, JsonSchema)]
#[serde(deny_unknown_fields)]
pub struct McpAttributionSource {
#[serde(default, skip_serializing_if = "Option::is_none")]
pub connector_id: Option<String>,
#[serde(default, skip_serializing_if = "Option::is_none")]
pub plugin_id: Option<String>,
pub server_name: String,
pub tool_name: String,
pub first_turn_id: String,
}
/// Cumulative, model-invisible attribution reported on Responses API requests.
/// This is provenance data, not a training-eligibility decision or attestation.
#[derive(Clone, Debug, Default, PartialEq, Eq, Serialize, Deserialize, JsonSchema)]
#[serde(deny_unknown_fields)]
pub struct McpAttribution {
pub status: McpAttributionStatus,
/// First observed attribution error, absent for non-error states.
#[serde(
default,
skip_serializing_if = "Option::is_none",
deserialize_with = "deserialize_mcp_attribution_error_reason"
)]
pub error_reason: Option<McpAttributionErrorReason>,
#[serde(default, skip_serializing_if = "Vec::is_empty")]
pub sources: Vec<McpAttributionSource>,
}
fn deserialize_mcp_attribution_error_reason<'de, D>(
deserializer: D,
) -> Result<Option<McpAttributionErrorReason>, D::Error>
where
D: Deserializer<'de>,
{
let value = serde_json::Value::deserialize(deserializer)?;
Ok(serde_json::from_value(value).ok())
}
/// Bounded app-resource provenance retained across a compaction checkpoint.
#[derive(Clone, Debug, Default, PartialEq, Eq, Serialize, Deserialize, JsonSchema)]
pub struct McpResourceOriginCheckpoint {