verifiers 0.2.2.dev35__py3-none-any.whl → 0.2.2.dev37__py3-none-any.whl

verifiers/v1/harnesses/codex/harness.py

@@ -4,9 +4,12 @@ The static musl binary runs in Linux containers without additional runtime depen
 """
 
 import base64
+import hashlib
+import json
 import logging
 import re
 import shlex
+from collections import Counter
 
 from verifiers.v1.clients import ModelContext
 from verifiers.v1.configs.harness import HarnessConfig
@@ -47,7 +50,7 @@ class CodexHarnessConfig(HarnessConfig):
 
 class CodexHarness(Harness[CodexHarnessConfig]):
     APPENDS_SYSTEM_PROMPT = False  # TODO
-    SUPPORTS_MCP = False  # TODO
+    SUPPORTS_MCP = True
     SUPPORTS_RESUME = True
     SUPPORTS_SKILLS = True
 
@@ -128,7 +131,7 @@ class CodexHarness(Harness[CodexHarnessConfig]):
         argv = [
             CODEX_BIN,
             "exec",
-            *self._config_args(ctx, endpoint),
+            *self._config_args(ctx, endpoint, mcp_urls),
             *image_args,
             "--",
             prompt,
@@ -196,7 +199,7 @@ class CodexHarness(Harness[CodexHarnessConfig]):
             "exec",
             "resume",
             "--last",
-            *self._config_args(ctx, endpoint),
+            *self._config_args(ctx, endpoint, mcp_urls),
             "--",
             "\n\n".join(texts),
         ]
@@ -228,7 +231,12 @@ class CodexHarness(Harness[CodexHarnessConfig]):
             "CODEX_HOME": home,
         }
 
-    def _config_args(self, ctx: ModelContext, endpoint: str) -> list[str]:
+    def _config_args(
+        self,
+        ctx: ModelContext,
+        endpoint: str,
+        mcp_urls: dict[str, str],
+    ) -> list[str]:
         # Values are Codex feature names such as `shell_tool`; Codex owns validation.
         # https://developers.openai.com/codex/config-reference#features
         tool_config = [
@@ -236,6 +244,49 @@ class CodexHarness(Harness[CodexHarnessConfig]):
             for tool in self.config.disabled_tools or []
             for arg in ("--disable", tool)
         ]
+        # Set the whole table so dots in quoted server IDs are not interpreted as
+        # `-c` path separators.
+        mcp_config = (
+            [
+                "-c",
+                "mcp_servers={"
+                + ",".join(
+                    (
+                        f"{json.dumps(name, ensure_ascii=False)}="
+                        f"{{url={json.dumps(url, ensure_ascii=False)},required=true,"
+                        "startup_timeout_sec=60.0,tool_timeout_sec=600.0}"
+                    )
+                    for name, url in mcp_urls.items()
+                )
+                + "}",
+            ]
+            if mcp_urls
+            else []
+        )
+
+        # Codex keeps raw server IDs for MCP calls but normalizes the namespaces
+        # exposed to the model. Mirror rust-v0.144.5's per-character replacement,
+        # optional prefix, collision hash, and possible 49-character truncation.
+        namespace_bases: dict[str, str] = {}
+        for name in mcp_urls:
+            namespace = re.sub(r"[^a-zA-Z0-9_]", "_", name) or "_"
+            namespace_bases[name] = (
+                namespace if namespace.startswith("mcp__") else f"mcp__{namespace}"
+            )
+        namespace_counts = Counter(namespace_bases.values())
+        direct_mcp_namespaces: list[str] = []
+        for name, namespace in namespace_bases.items():
+            if namespace_counts[namespace] > 1:
+                suffix = hashlib.sha1(f"{name}\0{name}\0".encode()).hexdigest()[:12]
+                namespace = (
+                    f"{namespace[:-2]}_{suffix}__"
+                    if namespace.endswith("__")
+                    else f"{namespace}_{suffix}"
+                )
+            direct_mcp_namespaces.append(namespace)
+            if len(namespace) > 49:
+                direct_mcp_namespaces.append(namespace[:49])
+        direct_mcp_namespaces = list(dict.fromkeys(direct_mcp_namespaces))
         # `-c` values parse as TOML, falling back to a raw string (so the url / `responses`
         # come through literally); `requires_openai_auth=false` parses as a bool.
         return [
@@ -266,4 +317,16 @@ class CodexHarness(Harness[CodexHarnessConfig]):
             "-c",
             f"model_providers.{PROVIDER}.requires_openai_auth=false",
             *tool_config,
+            *mcp_config,
+            *(
+                [
+                    "-c",
+                    (
+                        "features.code_mode.direct_only_tool_namespaces="
+                        f"{json.dumps(direct_mcp_namespaces)}"
+                    ),
+                ]
+                if direct_mcp_namespaces
+                else []
+            ),
         ]

{verifiers-0.2.2.dev35.dist-info → verifiers-0.2.2.dev37.dist-info}/METADATA

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: verifiers
-Version: 0.2.2.dev35
+Version: 0.2.2.dev37
 Summary: Verifiers: Environments for LLM Reinforcement Learning
 Project-URL: Homepage, https://github.com/primeintellect-ai/verifiers
 Project-URL: Documentation, https://github.com/primeintellect-ai/verifiers

{verifiers-0.2.2.dev35.dist-info → verifiers-0.2.2.dev37.dist-info}/RECORD

@@ -256,7 +256,7 @@ verifiers/v1/harnesses/bash/program.py,sha256=c9XgZwlsDYAwrgLZ6CDbZefQ1W4EF-vjM-
 verifiers/v1/harnesses/claude_code/__init__.py,sha256=JkZQylMSA3o1fvBx_NM41y_umuahmOSqFjTrJ2hryTw,171
 verifiers/v1/harnesses/claude_code/harness.py,sha256=98k5T8d8nFQmxIU-ytkrkrrP6UNR5NxoGhVJRLVa-fI,4015
 verifiers/v1/harnesses/codex/__init__.py,sha256=ocyBvlpSO8c3pT6D1ebrQohoXYESQcCyPT5kN3j1-N4,132
-verifiers/v1/harnesses/codex/harness.py,sha256=m3WqlQF8aGa3XwZ_jWpVchmQuMNtfXyOZdo9P9xNqRs,10665
+verifiers/v1/harnesses/codex/harness.py,sha256=OG1riQdVmZzVpHENsd0UzsIbdIAxDMHgRikmo73HL5M,13071
 verifiers/v1/harnesses/kimi_code/__init__.py,sha256=1BEotSfuzmVcE7UrxHeAREM5d7J20cGkonHuowlrzu4,161
 verifiers/v1/harnesses/kimi_code/harness.py,sha256=Pc2uiBsia-uQFuXvaLMFdblE6TS-p0E2cvB6ofRrHK0,4515
 verifiers/v1/harnesses/mini_swe_agent/__init__.py,sha256=1OBu2Y-XqL72-dGhFdgB6sr5uX_peEEXT9vEjPy1ovU,182
@@ -327,8 +327,8 @@ verifiers/v1/utils/logging.py,sha256=OcMHA6NsYux3oIzjPuI95rDWmFBHNcHDjZeNIhXTX-Y
 verifiers/v1/utils/memory.py,sha256=ZkIvGk6uITAH5sKon65LifKPbvZr8mJ__PVc23FZpOQ,1835
 verifiers/v1/utils/sampling.py,sha256=JczGzBn6s3wsIrSY2Hy3hE8m6NreNgsNzhSjDikQqX0,1037
 verifiers/v1/utils/version.py,sha256=-obEo_-l9-D8FLef4hYxncOe-uJpxrM1g2Hig_37Sgs,1607
-verifiers-0.2.2.dev35.dist-info/METADATA,sha256=C_9jwe-5-02Xq4rt1sM686TyE0ZTlQh7VFRi898xpbI,4540
-verifiers-0.2.2.dev35.dist-info/WHEEL,sha256=lCkmxWfQsSc9CfIClYeavTdQeEX2toPqufh9gI35EQA,87
-verifiers-0.2.2.dev35.dist-info/entry_points.txt,sha256=dF82JUYEFslR1AOW8LZfuBWeZeQoyX4wCRrduklg8-I,551
-verifiers-0.2.2.dev35.dist-info/licenses/LICENSE,sha256=v0RrUsdV3IDoZhrRce297IXS3xMHNJ-_LdLpFAUWb9k,1072
-verifiers-0.2.2.dev35.dist-info/RECORD,,
+verifiers-0.2.2.dev37.dist-info/METADATA,sha256=UXHagxu-19jjKUKZ4hFZhSL4eOZoFV4kBwD-0MmsHOw,4540
+verifiers-0.2.2.dev37.dist-info/WHEEL,sha256=lCkmxWfQsSc9CfIClYeavTdQeEX2toPqufh9gI35EQA,87
+verifiers-0.2.2.dev37.dist-info/entry_points.txt,sha256=dF82JUYEFslR1AOW8LZfuBWeZeQoyX4wCRrduklg8-I,551
+verifiers-0.2.2.dev37.dist-info/licenses/LICENSE,sha256=v0RrUsdV3IDoZhrRce297IXS3xMHNJ-_LdLpFAUWb9k,1072
+verifiers-0.2.2.dev37.dist-info/RECORD,,