litellm/tests/e2e/mcp/mcp_client.py
mubashir1osmani ac5b51253a
test(e2e): add Other suite and Guardrails coverage incl. an MCP tool-call guardrail (#34149)
* test(e2e): add other suite covering master-key auth and health lifecycle

Covers the other.* holding-pen cells that were uncovered: master-key
valid_allows/invalid_denied on the admin /user/list gate, and the
lifecycle probes liveness.ping, readiness.public_probe,
readiness.reports_db_status, and readiness_details.authenticated_diagnostics.

New tests/e2e/other/ suite on the shared ProxyClient; the health probes
send no auth header to prove the public routes need no credential, and the
details route is asserted to reject an anonymous caller while exposing
version/db diagnostics to the master key.

* test(e2e): cover block_code_execution and openai_moderation guardrails

Extends the guardrails suite with two built-in guardrails registered per
request (default_on=False, opted in via the chat body's guardrails selector)
so neither intercepts unrelated traffic on the shared proxy.

block_code_execution.pre_call.blocks: a python code block plus a run-this
request is intercepted with the canned content-blocked message and the model
never runs, while the same code block asked about with don't-run-it reaches
the model. Verified live.

openai_moderations.pre_call.blocks: a flagged prompt is rejected 400 naming
the moderation policy while a benign prompt passes. The guardrail calls
OpenAI's moderation API; verifying it needs an OpenAI key with moderation
quota (this account currently 429s the moderation endpoint).

Adds a shared create_backend_model helper and a generic register() plus
per-request guardrails/max_tokens on the client so more built-ins can reuse
the same path.

* test(e2e): cover presidio PII masking (pre_call + post_call)

Registers a presidio guardrail per request (default_on=False) with the
analyzer/anonymizer bases supplied in the registration params, so the test
controls its own dependency and needs no proxy restart.

presidio.pre_call.masks: a repeat-verbatim request comes back with the
<EMAIL_ADDRESS> placeholder and never the raw email, proving the prompt was
anonymized before the model saw it.

presidio.post_call.masks: with apply_to_output the model's own emitted email
is masked on the way out, so the caller never receives the raw value.

Both verified live against real presidio analyzer + anonymizer containers.
logging_only is intentionally not covered: /spend/logs exposes no prompt
messages to read back the masked log, and a logging_only run also masked the
response, contradicting its contract; noted in the module docstring for a
follow-up.

* test(e2e): cover presidio logging_only masking via OTEL read-back

Adds the third presidio cell, guardrail.presidio.logging_only.masks. The
logging_only contract (mask what is logged, do not block) is verified by
reading the request's gen-AI span back from the real OTEL destination: the
span's gen_ai.input.messages attribute carries the <EMAIL_ADDRESS> placeholder,
never the raw email, and the call itself is not blocked.

Reads the trace via the shared OtelReader, promoted from logging/ to the suite
root so both suites use it. The masked prompt is polled to a deadline because
logging_only masks the payload asynchronously and the span can briefly export
before the mask lands. Drops the throwaway chat_send in favor of the existing
transport.send for the call-id capture.

* fix(e2e): tolerate cross-pod guardrail sync delay in team-opt-out test

Stage runs multiple gateway pods behind the shared key. POST /guardrails
registers a new default-on guardrail in-process immediately only on the
pod that served the create call; every other pod picks it up on its next
periodic DB sync (proxy_server.py, every 30s), so the very next chat call
can race a pod that has not synced yet. Poll to a 40s deadline instead of
asserting on the first response, matching the existing pattern in
test_budget_reset_advances_e2e.py.

* test(e2e): cover a guardrail on the MCP tool-call path (content_filter pre_mcp_call)

Adds guardrail.litellm_content_filter.pre_mcp_call.blocks: against the real
Datadog MCP server, a content_filter guardrail configured mode=pre_mcp_call
blocks a banned keyword in an MCP tool call's arguments with HTTP 400 attributed
to the pre_mcp_call hook, and lets a clean argument reach the upstream server.

The guardrail attaches with default_on because per-key/request guardrail
selection is dropped from the synthetic MCP request the hook sees; the banned
keyword is unique per run so default_on only intercepts this test's own call.
mode must be pre_mcp_call - a pre_call config silently no-ops on tools/call
because the event type is rewritten for call_mcp_tool.

Drives the tool directly via /mcp-rest/tools/call for a deterministic check of
the same pre_mcp_call enforcement the OpenAI-SDK chat path hits when a model
invokes an MCP tool.

* fix(e2e): mid-conversation messages test uses client.proxy not client.gateway

EndpointsClient exposes .proxy after the Gateway->ProxyClient rename; the
mid-conversation system test still referenced .gateway, which fails the e2e
basedpyright gate. Aligns it with the rest of the harness.

* test(e2e): address review on the guardrail coverage

MCP tool-call guardrail: poll the banned call until the guardrail is enforced
instead of asserting on the first call, so the control-plane -> data-plane
guardrail sync cannot race the check into a false pass-through; add a repeat
banned call after enforcement to guard against a partial-propagation state.

OpenAI moderation: distinguish a moderation-endpoint 429 (rate limit / no
moderation quota) from a guardrail failure, so an account-capability gap reads
as such rather than as "did not block". Runs green with a moderation-capable key.

* test(e2e): close partial-propagation false-pass in MCP guardrail block test

The single post-block repeat call could be load-balanced back to the same
already-synced data-plane pod, so the test could pass while another pod still
lacked the guardrail and let the banned MCP call reach Datadog. Anchor a wait to
the guardrail create time (every pod is guaranteed to have DB-synced only after a
full ~30s sync interval), then require the banned call to stay blocked across
several attempts; a pass-through after that window is a real leak, not a race.

* test(e2e): drop xfail-style rate-limit branch from openai_moderation test

OpenAI's /v1/moderations is free and returns 200 with the env key (verified
directly), so the RateLimitedError branch mislabeled the failure: a 429 there is
insufficient_quota (no account billing), not throttling. The branch also only
printed a softer message before failing anyway, an xfail-in-disguise the e2e rules
forbid. A 429 now falls through and fails loudly with the full result.
2026-07-21 14:06:29 -07:00

268 lines
8.1 KiB
Python

"""Client for the MCP e2e suite: admin server registration plus the api_key tool
surface.
An admin registers an upstream MCP server through the management API
(`/v1/mcp/server`, persisted in the DB) and grants a virtual key access to it via
`object_permission.mcp_servers`. Keys then reach the server through the REST bridge
the proxy exposes for api_key auth (`/mcp-rest/tools/list`, `/mcp-rest/tools/call`),
which `user_api_key_auth` gates the same way the JSON-RPC `/mcp` surface does. The
request/response bodies are co-located here because only this suite speaks MCP.
"""
from __future__ import annotations
from collections.abc import Mapping
from dataclasses import dataclass
from pydantic import BaseModel, ConfigDict, Field, RootModel
from e2e_http import Headers, NoBody, Result, unwrap
from models import KeyGenerateBody, ObjectPermission
from proxy_client import ProxyClient
McpToolArg = str | int | float | bool | list[str] | dict[str, str]
McpToolArguments = Mapping[str, McpToolArg]
class ApiKeyHeaders(Headers):
x_litellm_api_key: str = Field(serialization_alias="x-litellm-api-key")
class McpServerNewBody(BaseModel):
server_name: str
alias: str
url: str
transport: str = "http"
auth_type: str | None = None
static_headers: dict[str, str] | None = None
allowed_tools: list[str] | None = None
class McpServerNewResponse(BaseModel):
server_id: str
class McpServerRow(BaseModel):
server_id: str
alias: str | None = None
url: str | None = None
class McpServersListResponse(RootModel[list[McpServerRow]]):
pass
class McpToolMcpInfo(BaseModel):
server_id: str | None = None
alias: str | None = None
class McpToolEntry(BaseModel):
name: str
description: str | None = None
mcp_info: McpToolMcpInfo | None = None
class McpToolsListResponse(BaseModel):
tools: list[McpToolEntry] = []
error: str | None = None
message: str | None = None
def tool_names_for_server(self, server_id: str) -> frozenset[str]:
return frozenset(
tool.name
for tool in self.tools
if tool.mcp_info is not None and tool.mcp_info.server_id == server_id
)
def tool_name_containing(self, server_id: str, needle: str) -> str | None:
needle_l = needle.lower()
for tool in self.tools:
if tool.mcp_info is None or tool.mcp_info.server_id != server_id:
continue
if needle_l in tool.name.lower() or tool.name.lower().endswith(needle_l):
return tool.name
return None
class BlockedWordSpec(BaseModel):
keyword: str
action: str = "BLOCK"
class ContentFilterMcpParams(BaseModel):
"""litellm_content_filter params scoped to the MCP tool-call hook. mode is
pre_mcp_call because a pre_call config silently no-ops on the tools/call path
(the event type is rewritten to pre_mcp_call for call_mcp_tool), and default_on
is required there because per-key/request guardrail selection is dropped from
the synthetic MCP request the hook sees."""
guardrail: str = "litellm_content_filter"
mode: str = "pre_mcp_call"
default_on: bool = True
blocked_words: list[BlockedWordSpec]
class GuardrailSpecBody(BaseModel):
guardrail_name: str
litellm_params: ContentFilterMcpParams
class GuardrailCreateBody(BaseModel):
guardrail: GuardrailSpecBody
class GuardrailCreateResponse(BaseModel):
guardrail_id: str
class McpCallToolBody(BaseModel):
name: str
arguments: dict[str, McpToolArg]
server_id: str
class McpCallContent(BaseModel):
type: str | None = None
text: str | None = None
class McpCallToolResponse(BaseModel):
model_config = ConfigDict(populate_by_name=True)
content: list[McpCallContent] = []
is_error: bool | None = Field(default=None, alias="isError")
@property
def first_text(self) -> str | None:
return self.content[0].text if self.content else None
@property
def all_text(self) -> str:
return "\n".join(part.text for part in self.content if part.text)
@dataclass(frozen=True, slots=True)
class McpClient:
proxy: ProxyClient
def register_server(
self,
*,
server_name: str,
alias: str,
url: str,
transport: str = "http",
auth_type: str | None = None,
static_headers: dict[str, str] | None = None,
allowed_tools: list[str] | None = None,
) -> str:
return unwrap(
self.proxy.transport.post(
"/v1/mcp/server",
headers=self.proxy.transport.master,
json=McpServerNewBody(
server_name=server_name,
alias=alias,
url=url,
transport=transport,
auth_type=auth_type,
static_headers=static_headers,
allowed_tools=allowed_tools,
),
response_type=McpServerNewResponse,
)
).server_id
def delete_server(self, server_id: str) -> None:
_ = self.proxy.transport.delete(
f"/v1/mcp/server/{server_id}",
headers=self.proxy.transport.master,
json=NoBody(),
response_type=NoBody,
)
def registered_servers(self) -> list[McpServerRow]:
return unwrap(
self.proxy.transport.get(
"/v1/mcp/server",
headers=self.proxy.transport.master,
params=NoBody(),
response_type=McpServersListResponse,
)
).root
def generate_key(
self,
*,
user_id: str,
mcp_servers: list[str] | None,
models: list[str] | None = None,
) -> str:
object_permission = (
ObjectPermission(mcp_servers=mcp_servers) if mcp_servers is not None else None
)
return self.proxy.generate_key(
KeyGenerateBody(
models=models if models is not None else [],
user_id=user_id,
object_permission=object_permission,
)
)
def list_tools(self, key: str) -> Result[McpToolsListResponse]:
return self.proxy.transport.get(
"/mcp-rest/tools/list",
headers=ApiKeyHeaders(x_litellm_api_key=key),
params=NoBody(),
response_type=McpToolsListResponse,
)
def register_mcp_content_filter(self, *, name: str, blocked_keyword: str) -> str:
"""Register a default-on content-filter guardrail that runs on the MCP
tool-call hook (pre_mcp_call) and blocks a single keyword. The keyword is
unique per test, so default_on only ever intercepts this test's own
banned tool call on the shared proxy."""
return unwrap(
self.proxy.transport.post(
"/guardrails",
headers=self.proxy.transport.master,
json=GuardrailCreateBody(
guardrail=GuardrailSpecBody(
guardrail_name=name,
litellm_params=ContentFilterMcpParams(
blocked_words=[BlockedWordSpec(keyword=blocked_keyword)],
),
)
),
response_type=GuardrailCreateResponse,
)
).guardrail_id
def delete_guardrail(self, guardrail_id: str) -> None:
_ = self.proxy.transport.delete(
f"/guardrails/{guardrail_id}",
headers=self.proxy.transport.master,
json=NoBody(),
response_type=NoBody,
)
def call_tool(
self,
key: str,
*,
server_id: str,
name: str,
arguments: McpToolArguments,
) -> Result[McpCallToolResponse]:
return self.proxy.transport.post(
"/mcp-rest/tools/call",
headers=ApiKeyHeaders(x_litellm_api_key=key),
json=McpCallToolBody(
name=name, arguments=dict(arguments), server_id=server_id
),
response_type=McpCallToolResponse,
)
def build_client(proxy: ProxyClient) -> McpClient:
return McpClient(proxy=proxy)