青榆牧 a94b2d8897
feat(mcp): map request-scoped secrets to MCP HTTP/SSE headers (#5010)
* feat(mcp): map request-scoped secrets to HTTP/SSE headers

`user_auth` binds a credential to a configured DeerFlow user, so a caller
that picks the credential per request — a multi-tenant gateway, a per-run
API key, one shared MCP server fronting several environments — had to
register one MCP server entry per credential.

Add a declarative `mcpServers.<server>.headers_from_context` block mapping
HTTP header names to keys of the run request's `config.context.secrets`
carrier. A new built-in interceptor resolves the mapping on every tool call
and rewrites those headers, mirroring `user_scoped_auth`. The config file
stores names only, never a credential, so the Gateway returns the block
unmasked.

Registered after OAuth and `user_auth` in the interceptor chain: the later
interceptor runs closer to the transport, and the value chosen for this one
request is the most specific, so it wins. Fail-closed by default — a mapped
key missing from the request raises a `ToolException` naming only that key,
because falling back to the server's discovery credential would send one
tenant's call under another tenant's authority. `on_missing: "passthrough"`
opts out.

Durable background tasks are excluded: `McpTaskToolCaller` drives status and
cancel polls after the Agent run ends, where no run context exists, so the
fail-closed interceptor would deny every poll. Those calls keep using
server-level credentials, and a server declaring both `headers_from_context`
and `task_toolsets` now logs a warning.

Also corrects the custom-interceptor example in docs/MCP_SERVER.md (and the
matching claim in skills/AGENTS.md), which read request secrets from
`langgraph.config.get_config()["context"]`. That key is `None` inside a tool
call — the run context rides the LangGraph runtime, not the RunnableConfig
propagated to child runnables — so interceptors written from that example
never saw a value. The example now reads `request.runtime`, and
tests/test_mcp_context_headers.py pins LangGraph's runtime-injection rule by
driving a real langchain-mcp-adapters tool through a real graph with the
ambient-runtime fallback disabled.

Closes #5005

Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>

* fix(mcp): resolve credential headers case-insensitively, carry them on durable submit

Review follow-ups on `headers_from_context`.

HTTP field names are case-insensitive, but every dict on the path to the wire
is not: `build_server_params` copies the operator's static `headers` spelling
verbatim, and langchain-mcp-adapters merges interceptor overrides into the
connection with a plain `{**connection_headers, **override_headers}` splat. A
static `authorization` and an injected `Authorization` therefore both reached
httpx as separate field lines, and a server reading the field with a
single-value accessor got the static discovery credential — inverting the
documented `headers` < `oauth` < `user_auth` < `headers_from_context`
precedence and running a per-request call under the shared credential.

Normalizing inside the interceptor cannot fix that on its own: the adapter
builds the request with `headers=None`, so an interceptor never sees the
connection's static headers and cannot displace them however it spells its own
key. A new `mcp/headers.py::apply_header_overrides` therefore drops any key
differing only in case and emits the spelling the connection already uses.
Applied to `headers_from_context`, `user_auth`, the OAuth interceptor, the
OAuth discovery-header write, and the durable-task connection merge, which all
carried the same collision. `headers_from_context.headers` now also rejects one
header mapped under two spellings at config load, in both the harness model and
the Gateway mirror.

Durable submit now carries the mapped headers, as docs/MCP_SERVER.md already
promised. `McpTaskToolCaller` disabled the interceptor for the whole caller, but
that caller serves submit as well as the polls, and submit is awaited inline
inside the Agent's tool call — where the run's LangGraph runtime is still the
ambient contextvar, so no secret has to be threaded through `TaskSubmitRequest`
or reach durable storage. The caller builds one chain and keeps a second view of
it without the context-headers interceptor; `call_tool` takes
`request_scoped_headers`, set only by `OrdinaryMcpTaskDriver.submit`. Status and
cancel keep server-level credentials, so background polls still cannot fail
closed, and the startup warning now describes the half it actually covers.

`_merge_preserving_secrets` restores masked extras inside `headers_from_context`
instead of writing the `***` sentinel back over the stored value, matching the
treatment `user_auth` extras and server-level extras already get; extras a PUT
omits carry over as well, while the declared mapping still replaces verbatim so
a round trip can remove an entry. `extra="allow"` plus name-based sensitivity
detection means the usual casualty is a name-valued key such as `tokenHeader`,
not only a credential.

The existing override test seeded the static header onto `request.headers`,
which production never does, so it modelled a merge that really happens one
layer down; the new tests drive a real adapter tool through a real connection
and assert on the headers the session is opened with, and the durable-submit
test runs through a real tool node with no runtime patching.

Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>

* fix(mcp): reject case-insensitive duplicate static header names

* fix(mcp): preserve omitted headers_from_context fields on partial updates

---------

Co-authored-by: Claude Opus 5 <noreply@anthropic.com>
2026-08-27 10:42:42 +08:00

127 lines
6.0 KiB
Python

"""Per-user credential injection for shared MCP servers.
One configured HTTP/SSE MCP server can serve several DeerFlow users, each
authenticated to the remote service with their own credential. A server opts in
by declaring a ``user_auth`` block (:class:`McpUserScopedAuthConfig`) mapping
DeerFlow user ids to credential header values. On every tool call the
interceptor resolves the authenticated user and rewrites the configured header
via ``request.override(headers=...)`` — the same per-call mechanism the OAuth
interceptor uses.
The server entry's static ``headers`` are used only for startup tool discovery
(``tools/list``); they never authenticate a user's tool call when ``user_auth``
is enabled for that server, except under an explicit ``on_missing:
"passthrough"`` opt-out.
Fail-closed by default: an unmapped user (including the anonymous
``DEFAULT_USER_ID`` fallback), or a mapped credential whose ``$ENV_VAR``
reference resolved to an empty string, gets an actionable ``ToolException``
instead of another user's credential or the discovery credential.
"""
from __future__ import annotations
import logging
from typing import Any
from langchain_core.tools import ToolException
from deerflow.config.extensions_config import ExtensionsConfig, McpUserScopedAuthConfig
from deerflow.mcp.headers import apply_header_overrides, header_spellings
from deerflow.runtime.user_context import resolve_runtime_user_id
logger = logging.getLogger(__name__)
def _current_runtime() -> Any | None:
"""Best-effort access to the LangGraph runtime for the current tool call.
``get_runtime()`` raises outside a runtime context (embedded clients, unit
tests, discovery paths); ``resolve_runtime_user_id`` accepts ``None`` and
falls back to LangGraph auth config and the request-scoped user
ContextVar, so failures here reduce accuracy but never crash the call.
"""
try:
from langgraph.runtime import get_runtime
return get_runtime()
except Exception:
return None
def build_user_scoped_auth_interceptor(extensions_config: ExtensionsConfig) -> Any | None:
"""Build a tool interceptor injecting per-user credentials, or ``None``.
Returns ``None`` when no enabled server declares an enabled ``user_auth``
block, so callers can skip registration entirely (mirrors
``build_oauth_tool_interceptor``).
"""
user_auth_by_server: dict[str, McpUserScopedAuthConfig] = {}
# The server's static header spellings, so a configured ``header`` that
# differs from the static one only in case still *replaces* it at the
# adapter's case-sensitive connection merge (see ``mcp/headers.py``).
spellings_by_server: dict[str, dict[str, str]] = {}
for server_name, server_config in extensions_config.get_enabled_mcp_servers().items():
if server_config.user_auth is None or not server_config.user_auth.enabled:
continue
if server_config.type not in ("sse", "http"):
# A stdio server has no HTTP headers: the pooled stdio path forwards
# rewritten headers as call meta, never a transport header, so the
# credential would go nowhere while deny errors still fired for
# unmapped users. Warn-and-skip matches the existing convention for
# transport/config mismatches (e.g. tool_call_timeout on non-stdio).
logger.warning(
"MCP server '%s' declares user_auth but uses the '%s' transport; user-scoped credentials only apply to 'sse'/'http' servers — ignoring user_auth for this server",
server_name,
server_config.type,
)
continue
user_auth_by_server[server_name] = server_config.user_auth
spellings_by_server[server_name] = header_spellings(server_config.headers)
if not user_auth_by_server:
return None
async def user_scoped_auth_interceptor(request: Any, handler: Any) -> Any:
user_auth = user_auth_by_server.get(request.server_name)
if user_auth is None:
return await handler(request)
# Prefer the runtime attached to the request (set by the adapter when
# the call originates inside a graph); fall back to the ambient
# LangGraph runtime, then to resolve_runtime_user_id's own chain
# (LangGraph auth config → request-scoped user ContextVar → default).
runtime = getattr(request, "runtime", None)
if runtime is None:
runtime = _current_runtime()
user_id = resolve_runtime_user_id(runtime)
# Empty string covers a `$ENV_VAR` reference whose variable was unset:
# ExtensionsConfig.resolve_env_variables stores "" for those, and an
# empty credential must fail closed rather than send an empty header.
credential = user_auth.users.get(user_id, "")
if not credential:
if user_auth.on_missing == "passthrough":
return await handler(request)
logger.warning(
"Denied MCP tool call to server '%s': no user-scoped credential for user '%s'",
request.server_name,
user_id,
)
# The resolved id is included so the operator can copy the exact
# ``users`` key: it differs by deployment path (a safe-slug like
# ``alice-example-com-ab12cd34`` via LangGraph auth, a raw user
# UUID via the embedded Gateway). It is the caller's own id, so
# surfacing it leaks nothing across users.
raise ToolException(
f"No credential is configured for your account (user id '{user_id}') on MCP server '{request.server_name}'. Ask the operator to add this exact id to that server's user_auth.users map (or set its environment variable)."
)
updated_headers = apply_header_overrides(
request.headers,
{user_auth.header: credential},
spellings=spellings_by_server.get(request.server_name),
)
return await handler(request.override(headers=updated_headers))
return user_scoped_auth_interceptor