diff --git a/litellm/proxy/_experimental/mcp_server/outbound_credentials/session_credentials.py b/litellm/proxy/_experimental/mcp_server/outbound_credentials/session_credentials.py new file mode 100644 index 00000000000..08d5cc8b1f1 --- /dev/null +++ b/litellm/proxy/_experimental/mcp_server/outbound_credentials/session_credentials.py @@ -0,0 +1,190 @@ +"""Producer and consumer helpers for the gateway-level DCR session token. + +The aggregate ``/mcp`` front door (``mcp_gateway_dcr``) issues the identity-only session +tokens defined in :mod:`.session_token`. The gateway token endpoint mints them (producer) +after SSO sign-in, and at the MCP admission edge the gateway derives the session signing +key from the proxy ``master_key``, opens the bearer, and admits the request under the +recovered litellm user (consumer), reloading the live user record and policy before +anything runs. This module is the pure surface for both sides; the token-endpoint and +admission wiring live in their respective call sites. + +The signing key is derived with the same memory-hard scrypt construction as +:func:`~.bridge_credentials.envelope_keys_from_master_key` but under a distinct domain +label, so session tokens and bridge envelopes never share key material: a token of one +family is unverifiable in the other by key separation, on top of the distinct issuers, +prefixes, and claim shapes. +""" + +import hashlib +from datetime import datetime +from functools import lru_cache +from typing import Literal, TypeAlias + +from pydantic import BaseModel, ConfigDict, SecretStr + +from litellm.proxy._experimental.mcp_server.outbound_credentials.session_token import ( + OpenedSessionToken, + SessionExpired, + SessionKeys, + SessionPrincipal, + is_session_refresh_token, + is_session_token, + open_session_refresh_token, + open_session_token, +) + +_SESSION_SIGNING_KEY_DOMAIN = b"litellm-mcp-gateway:session-signing:" + +# scrypt work factors (RFC 7914), identical to the envelope KDF: memory-hard so a captured +# session token is not a cheap offline oracle for the master key. +_SCRYPT_N = 2**15 +_SCRYPT_R = 8 +_SCRYPT_P = 1 +_SCRYPT_MAXMEM = 128 * _SCRYPT_N * _SCRYPT_R * _SCRYPT_P * 2 +_DERIVED_KEY_BYTES = 32 + + +@lru_cache(maxsize=8) +def session_keys_from_master_key(master_key: str) -> SessionKeys: + """Derive the session signing key from the proxy master key. + + A memory-hard scrypt KDF (RFC 7914) over a session-specific domain-label salt yields a + 256-bit subkey from the one secret, so the producer (mint) and consumer (open) agree on + the key without persisting any. The domain label differs from both envelope labels in + :mod:`.bridge_credentials`, so compromise or misuse of one token family never crosses + into the other. The result is cached (the master key is fixed for a process); rotating + ``master_key`` invalidates every outstanding session, which is the intended behavior + for a signing-key change. + """ + signing = hashlib.scrypt( + master_key.encode(), + salt=_SESSION_SIGNING_KEY_DOMAIN, + n=_SCRYPT_N, + r=_SCRYPT_R, + p=_SCRYPT_P, + maxmem=_SCRYPT_MAXMEM, + dklen=_DERIVED_KEY_BYTES, + ).hex() + return SessionKeys(signing_key=SecretStr(signing)) + + +class NotSessionBearer(BaseModel): + """The bearer is not session-shaped; admission continues on its normal path.""" + + model_config = ConfigDict(frozen=True) + tag: Literal["not_session_bearer"] = "not_session_bearer" + + +class SessionBearerAdmitted(BaseModel): + """A valid session access token: the principal to admit under after a live reload.""" + + model_config = ConfigDict(frozen=True) + tag: Literal["admitted"] = "admitted" + principal: SessionPrincipal + + +class SessionBearerInvalid(BaseModel): + """The bearer is session-shaped but must not admit (expired, tampered, wrong key, or a + refresh token presented at the tool-call edge); admission fails closed with the + ``invalid_token`` challenge rather than falling through to another arm. ``expired`` + distinguishes a routine expiry (debug-log worthy) from a tampered or foreign token.""" + + model_config = ConfigDict(frozen=True) + tag: Literal["invalid"] = "invalid" + expired: bool = False + + +SessionBearerResult: TypeAlias = NotSessionBearer | SessionBearerAdmitted | SessionBearerInvalid + + +def _strip_bearer(value: str) -> str: + parts = value.split(None, 1) + if len(parts) == 2 and parts[0].lower() == "bearer": + return parts[1] + return value + + +def is_session_bearer_shaped(authorization_value: str) -> bool: + """Cheap, keyless test that an ``Authorization`` value carries a session token of either + kind (optional ``Bearer`` scheme stripped). The admission edge engages the session arm + for an access token (to admit) and for a refresh token (to reject it explicitly, since + a refresh credential is never usable at the tool-call edge); anything else falls + through to normal admission.""" + candidate = _strip_bearer(authorization_value) + return is_session_token(candidate) or is_session_refresh_token(candidate) + + +def resolve_session_bearer( + authorization_value: str, + keys: SessionKeys, + now: datetime, +) -> SessionBearerResult: + """Classify an ``Authorization`` value presented at the aggregate MCP edge. + + Strips an optional ``Bearer`` scheme, then returns ``NotSessionBearer`` for a + non-session bearer (normal admission continues), ``SessionBearerAdmitted`` with the + recovered principal for a valid access token, and ``SessionBearerInvalid`` for a + session-shaped bearer that must not admit. Never raises: total over hostile input via + :func:`~.session_token.open_session_token`. + + A refresh token is ``SessionBearerInvalid`` here: it is a valid gateway credential but + only ever presented back to the token endpoint, so admission must fail it closed rather + than let it fall through to another arm. + """ + candidate = _strip_bearer(authorization_value) + if is_session_refresh_token(candidate): + return SessionBearerInvalid() + if not is_session_token(candidate): + return NotSessionBearer() + opened = open_session_token(candidate, keys, now) + if isinstance(opened, OpenedSessionToken): + return SessionBearerAdmitted(principal=opened.principal) + return SessionBearerInvalid(expired=isinstance(opened, SessionExpired)) + + +class SessionRefreshOpened(BaseModel): + """A valid session refresh token presented to the token endpoint: the principal to + re-validate and renew under.""" + + model_config = ConfigDict(frozen=True) + tag: Literal["opened"] = "opened" + principal: SessionPrincipal + + +class SessionRefreshInvalid(BaseModel): + """The presented refresh grant is not a valid session refresh token for this client + (not refresh-shaped, will not open, or bound to a different ``client_id``); the token + endpoint fails the refresh closed.""" + + model_config = ConfigDict(frozen=True) + tag: Literal["invalid"] = "invalid" + + +SessionRefreshResult: TypeAlias = SessionRefreshOpened | SessionRefreshInvalid + + +def open_session_refresh_bearer( + refresh_value: str, + keys: SessionKeys, + now: datetime, + expected_client_id: str, +) -> SessionRefreshResult: + """Open a session refresh token presented on a ``refresh_token`` grant. + + The token-endpoint mirror of :func:`resolve_session_bearer`: strips an optional + ``Bearer`` scheme, then returns ``SessionRefreshOpened`` with the recovered principal, + or ``SessionRefreshInvalid`` for anything that is not a valid session refresh token + issued to ``expected_client_id``. Never raises. The client binding (RFC 6749 section 6) + stops a refresh token stolen from one DCR client from being renewed through another; + ``client_id`` is not a secret (the caller presents it), so a plain equality check is + sufficient and, unlike ``hmac.compare_digest`` on ``str``, does not raise on non-ASCII. + """ + candidate = _strip_bearer(refresh_value) + if not is_session_refresh_token(candidate): + return SessionRefreshInvalid() + opened = open_session_refresh_token(candidate, keys, now) + if not isinstance(opened, OpenedSessionToken): + return SessionRefreshInvalid() + if opened.principal.client_id != expected_client_id: + return SessionRefreshInvalid() + return SessionRefreshOpened(principal=opened.principal) diff --git a/litellm/proxy/_experimental/mcp_server/outbound_credentials/session_token.py b/litellm/proxy/_experimental/mcp_server/outbound_credentials/session_token.py new file mode 100644 index 00000000000..78b1f7e4916 --- /dev/null +++ b/litellm/proxy/_experimental/mcp_server/outbound_credentials/session_token.py @@ -0,0 +1,356 @@ +"""Identity-only session tokens for the gateway-level (aggregate ``/mcp``) DCR front door. + +A DCR client that signs in through LiteLLM SSO holds ONE bearer that carries ONLY a +litellm identity; unlike the :mod:`.envelope` bridge bearer it seals no upstream +credential, because the custody model vaults every upstream token server-side in +``LiteLLM_MCPUserCredentials`` and egress resolves them by user at call time. The token +is therefore a stable REFERENCE, not an authorization: admission reloads the live user +record and policy on every request, so deactivating the user (or their team) kills +outstanding sessions immediately without a revocation store. + +Wire shape: ``llm_session_`` (access) / ``llm_srefresh_`` (refresh) + an HS256 JWT, +the same signing approach as :mod:`.envelope`. Claims are ``iss``/``iat``/``exp`` +plus ``kind``, ``user_id``, and ``client_id``; ``client_id`` binds the refresh token +to the DCR client it was issued to (RFC 6749 section 6) and is carried on the access +token for parity and audit. There is no encrypted payload: nothing in a session token +is secret beyond the signature, and reprs never print the signed value because minted +tokens are ``SecretStr``. + +This module is pure and unwired: it imports nothing from endpoint or edge code, reads +no proxy globals, and takes all key material and the clock as explicit parameters. +Failures are values: :func:`open_session_token` and :func:`open_session_refresh_token` +are total over hostile, attacker-controlled input and return a +``SessionTokenOpenError`` variant rather than raising. PyJWT's ``iat``/``nbf``/``exp`` +validators are disabled for the same reasons documented in :mod:`.envelope` (they +raise on hostile claim types and compare against the wall clock instead of the +injected ``now``); the strict pydantic claims model is the sole, total type gate. +""" + +from __future__ import annotations + +from datetime import datetime, timedelta +from typing import Literal, TypeAlias + +import jwt +from pydantic import BaseModel, ConfigDict, Field, SecretStr, ValidationError + +SESSION_TOKEN_PREFIX = "llm_session_" +"""Marker prefix on every serialized session ACCESS token so the admission edge can cheaply +tell a gateway session from a litellm key, JWT, or bridge envelope before doing any +cryptography. Distinct from the ``llm_env_``/``llm_refresh_`` envelope prefixes.""" + +SESSION_REFRESH_PREFIX = "llm_srefresh_" +"""Marker prefix on every serialized session REFRESH token. A distinct prefix keeps the two +credentials routable without crypto and, together with the signed ``kind`` claim, stops one +from being presented where the other is expected: the refresh token is only ever presented +back to the token endpoint, never at the MCP edge.""" + +SESSION_ISSUER = "litellm-mcp-gateway" +"""``iss`` claim stamped into every session token and required back on open. Distinct from +the envelope issuer so a token of one family can never validate in the other even under a +hypothetical shared signing key.""" + +SESSION_TTL_SECONDS = 3600 +"""Session ACCESS token lifetime (1h), matching the access-envelope and BYOK session bearer +windows: a client-held credential never outlives a bounded window, and each refresh +re-validates the live user before re-minting.""" + +SESSION_REFRESH_TTL_SECONDS = 1209600 +"""Session REFRESH token lifetime (14 days), matching the refresh-envelope bound. Each +renewal re-validates the sealed user against the live record (deactivation gates it) and +rotates the refresh token, so the practical bound is idle time, not a fixed session.""" + +MAX_SESSION_TOKEN_BYTES = 4096 +"""Size cap on the serialized token (prefix + JWT, in bytes) and on any candidate accepted +by the openers. Session claims are small; the only variable-length field is ``client_id`` +(a sealed DCR client record), and 4096 leaves ample headroom under common 8-16KB header +limits while bounding hostile input before JWT parsing.""" + +_SESSION_JWT_ALGORITHM = "HS256" + +SessionTokenKind = Literal["session", "session_refresh"] +"""Which credential a session token is. Stamped into the signed claims and required to match +on open, so a signature-valid token of one kind cannot be replayed as the other even if its +wire prefix is swapped (the prefix is not part of the signed payload; this claim is).""" + + +class SessionPrincipal(BaseModel): + """The litellm user a session token identifies and the DCR client it was issued to. + + ``user_id`` is the SSO-established litellm user subject, never a credential: admission + reloads the live user record by it, so current role, team, and revocation state are + enforced at use time rather than frozen at mint time. ``client_id`` is the (stateless, + gateway-sealed) DCR client identifier the token was issued to; the token endpoint + requires it to match on the refresh grant. + """ + + model_config = ConfigDict(frozen=True) + user_id: str = Field(min_length=1) + client_id: str = Field(min_length=1) + + +class SessionKeys(BaseModel): + """Injected key material: the HS256 signing key. + + ``signing_key`` must be at least 32 bytes: HS256's HMAC-SHA256 has a 256-bit security + level, RFC 7518 requires a key of at least that size, and a shorter key makes PyJWT + emit ``InsecureKeyLengthWarning``. + """ + + model_config = ConfigDict(frozen=True) + signing_key: SecretStr = Field(min_length=32) + + +class MintedSessionToken(BaseModel): + """A minted session token: the client-held bearer value and when it expires.""" + + model_config = ConfigDict(frozen=True) + token: SecretStr + expires_at: datetime + + +class OpenedSessionToken(BaseModel): + """A validated session token of either kind: the principal it was minted for.""" + + model_config = ConfigDict(frozen=True) + principal: SessionPrincipal + + +class SessionTokenTooLarge(BaseModel): + """The serialized token exceeded ``MAX_SESSION_TOKEN_BYTES``; carries sizes only. Only + reachable through an oversized ``client_id``, which registration should have bounded.""" + + model_config = ConfigDict(frozen=True) + tag: Literal["session_token_too_large"] = "session_token_too_large" + size_bytes: int + max_bytes: int + + +SessionTokenMintError: TypeAlias = SessionTokenTooLarge + + +class NotASessionToken(BaseModel): + """The candidate does not carry the expected session prefix.""" + + model_config = ConfigDict(frozen=True) + tag: Literal["not_a_session_token"] = "not_a_session_token" + + +class SessionBadSignature(BaseModel): + """The JWT signature does not verify under the provided signing key.""" + + model_config = ConfigDict(frozen=True) + tag: Literal["session_bad_signature"] = "session_bad_signature" + + +class SessionExpired(BaseModel): + """The token's ``exp`` is not in the future relative to the provided ``now``.""" + + model_config = ConfigDict(frozen=True) + tag: Literal["session_expired"] = "session_expired" + + +class SessionMalformed(BaseModel): + """The token is not a well-formed session token: undecodable JWT, wrong issuer, wrong + ``kind``, or missing/mistyped/extra claims.""" + + model_config = ConfigDict(frozen=True) + tag: Literal["session_malformed"] = "session_malformed" + + +SessionTokenOpenError: TypeAlias = NotASessionToken | SessionBadSignature | SessionExpired | SessionMalformed + + +class _SessionClaims(BaseModel): + """Decoded-claims boundary that pins the exact shape the mints emit. + + ``user_id``/``client_id`` mirror the ``min_length`` constraints of + :class:`SessionPrincipal` so any claim set that validates here also constructs a + principal, keeping the openers raise-free: a correctly signed JWT with an empty + identity claim fails here and maps to ``SessionMalformed``. ``strict`` rejects coerced + types (``exp: "123"``) and ``extra="forbid"`` rejects any claim the gateway never + mints; PyJWT's own registered-claim validators are disabled at decode (see module + docstring), so this model is the sole, total type gate for every claim. + """ + + model_config = ConfigDict(frozen=True, strict=True, extra="forbid") + iss: str + iat: int + exp: int + kind: SessionTokenKind + user_id: str = Field(min_length=1) + client_id: str = Field(min_length=1) + + +def is_session_token(candidate: str) -> bool: + """Cheap prefix check for a session ACCESS token so the admission edge can route gateway + sessions vs keys, JWTs, and envelopes without crypto.""" + return candidate.startswith(SESSION_TOKEN_PREFIX) + + +def is_session_refresh_token(candidate: str) -> bool: + """Cheap prefix check for a session REFRESH token so the token endpoint can route a + refresh grant without crypto.""" + return candidate.startswith(SESSION_REFRESH_PREFIX) + + +def mint_session_token( + principal: SessionPrincipal, + keys: SessionKeys, + now: datetime, +) -> MintedSessionToken | SessionTokenMintError: + """Mint the short-lived session ACCESS token for ``principal``. + + ``exp`` is ``SESSION_TTL_SECONDS`` from ``now``. Returns ``SessionTokenTooLarge`` when + the serialized token exceeds ``MAX_SESSION_TOKEN_BYTES``. + """ + return _mint( + kind="session", + prefix=SESSION_TOKEN_PREFIX, + principal=principal, + expires_at=now + timedelta(seconds=SESSION_TTL_SECONDS), + keys=keys, + now=now, + ) + + +def mint_session_refresh_token( + principal: SessionPrincipal, + keys: SessionKeys, + now: datetime, +) -> MintedSessionToken | SessionTokenMintError: + """Mint the long-lived session REFRESH token for ``principal``. + + ``exp`` is ``SESSION_REFRESH_TTL_SECONDS`` from ``now``. Minting a distinct + ``kind="session_refresh"`` claim is what keeps a refresh token from ever opening as an + access credential at the MCP edge. + """ + return _mint( + kind="session_refresh", + prefix=SESSION_REFRESH_PREFIX, + principal=principal, + expires_at=now + timedelta(seconds=SESSION_REFRESH_TTL_SECONDS), + keys=keys, + now=now, + ) + + +def open_session_token( + candidate: str, + keys: SessionKeys, + now: datetime, +) -> OpenedSessionToken | SessionTokenOpenError: + """Validate a session ACCESS ``candidate`` and recover the principal. + + Never raises for bad input: every invalid, expired, tampered, or wrong-kind candidate + maps to a distinct ``SessionTokenOpenError`` variant. + """ + return _open(candidate, prefix=SESSION_TOKEN_PREFIX, expected_kind="session", keys=keys, now=now) + + +def open_session_refresh_token( + candidate: str, + keys: SessionKeys, + now: datetime, +) -> OpenedSessionToken | SessionTokenOpenError: + """Validate a session REFRESH ``candidate`` and recover the principal. + + Total over hostile input exactly like :func:`open_session_token`. The + ``kind="session_refresh"`` claim is required, so an access token re-prefixed as a + refresh one is rejected as ``SessionMalformed``. + """ + return _open(candidate, prefix=SESSION_REFRESH_PREFIX, expected_kind="session_refresh", keys=keys, now=now) + + +def _mint( + kind: SessionTokenKind, + prefix: str, + principal: SessionPrincipal, + expires_at: datetime, + keys: SessionKeys, + now: datetime, +) -> MintedSessionToken | SessionTokenTooLarge: + """Sign the claims for either token kind and enforce the size cap. Shared by both mints + so the JWT shape, issuer, and size guard cannot drift between access and refresh.""" + claims = _SessionClaims( + iss=SESSION_ISSUER, + iat=int(now.timestamp()), + exp=int(expires_at.timestamp()), + kind=kind, + user_id=principal.user_id, + client_id=principal.client_id, + ) + token = prefix + jwt.encode( + claims.model_dump(), keys.signing_key.get_secret_value(), algorithm=_SESSION_JWT_ALGORITHM + ) + size_bytes = len(token.encode("utf-8")) + if size_bytes > MAX_SESSION_TOKEN_BYTES: + return SessionTokenTooLarge(size_bytes=size_bytes, max_bytes=MAX_SESSION_TOKEN_BYTES) + return MintedSessionToken(token=SecretStr(token), expires_at=expires_at) + + +def _open( + candidate: str, + prefix: str, + expected_kind: SessionTokenKind, + keys: SessionKeys, + now: datetime, +) -> OpenedSessionToken | SessionTokenOpenError: + """Prefix-route, size-bound, signature-verify, kind-check, and expiry-check an + attacker-controlled candidate, shared by both openers so the security gate is identical + for access and refresh. Returns the opened token or a distinct error; never raises.""" + if not candidate.startswith(prefix): + return NotASessionToken() + # UTF-8 byte length is never below character length, so a character count already over + # the cap rejects an oversize candidate in O(1) without encoding it; the exact byte + # check then runs only on candidates already bounded to the cap in characters. + if len(candidate) > MAX_SESSION_TOKEN_BYTES: + return SessionMalformed() + if len(candidate.encode("utf-8", "surrogatepass")) > MAX_SESSION_TOKEN_BYTES: + return SessionMalformed() + claims = _decode_claims(candidate.removeprefix(prefix), keys.signing_key) + if not isinstance(claims, _SessionClaims): + return claims + if claims.kind != expected_kind: + return SessionMalformed() + if now.timestamp() >= claims.exp: + return SessionExpired() + return OpenedSessionToken(principal=SessionPrincipal(user_id=claims.user_id, client_id=claims.client_id)) + + +def _decode_claims( + compact: str, + signing_key: SecretStr, +) -> _SessionClaims | SessionBadSignature | SessionMalformed: + """Verify the HS256 signature and shape of an attacker-controlled compact JWT. + + ``compact`` is fully hostile and bounded to ``MAX_SESSION_TOKEN_BYTES`` by the caller. + PyJWT's ``iat``/``nbf``/``exp`` validators are disabled: they raise on hostile claim + types and, for ``iat``/``nbf``, compare against the wall clock rather than the injected + ``now`` (``exp`` is checked by the caller against ``now``). Apart from a signature + mismatch, every decode failure is ``SessionMalformed``: a non-UTF-8 candidate surfaces + as ``UnicodeEncodeError`` (a ``ValueError``), a non-string registered claim as a + ``TypeError`` from PyJWT's claim validators, and a wrong issuer or structurally invalid + token as an ``InvalidTokenError``. ``_SessionClaims`` is the total type gate. + """ + try: + payload = jwt.decode( + compact, + signing_key.get_secret_value(), + algorithms=[_SESSION_JWT_ALGORITHM], + issuer=SESSION_ISSUER, + options={ + "verify_exp": False, + "verify_iat": False, + "verify_nbf": False, + "require": ["iss", "iat", "exp"], + }, + ) + except jwt.InvalidSignatureError: + return SessionBadSignature() + except (jwt.InvalidTokenError, ValueError, TypeError): + return SessionMalformed() + try: + return _SessionClaims.model_validate(payload) + except ValidationError: + return SessionMalformed() diff --git a/tests/test_litellm/proxy/_experimental/mcp_server/outbound_credentials/test_session_credentials.py b/tests/test_litellm/proxy/_experimental/mcp_server/outbound_credentials/test_session_credentials.py new file mode 100644 index 00000000000..8fa7c15d2d3 --- /dev/null +++ b/tests/test_litellm/proxy/_experimental/mcp_server/outbound_credentials/test_session_credentials.py @@ -0,0 +1,135 @@ +"""Tests for the session-token KDF and the edge/token-endpoint resolvers.""" + +from datetime import datetime, timedelta, timezone + +import pytest + +from litellm.proxy._experimental.mcp_server.outbound_credentials.bridge_credentials import ( + envelope_keys_from_master_key, +) +from litellm.proxy._experimental.mcp_server.outbound_credentials.session_credentials import ( + NotSessionBearer, + SessionBearerAdmitted, + SessionBearerInvalid, + SessionRefreshInvalid, + SessionRefreshOpened, + is_session_bearer_shaped, + open_session_refresh_bearer, + resolve_session_bearer, + session_keys_from_master_key, +) +from litellm.proxy._experimental.mcp_server.outbound_credentials.session_token import ( + SESSION_TTL_SECONDS, + MintedSessionToken, + SessionPrincipal, + mint_session_refresh_token, + mint_session_token, +) + +NOW = datetime(2026, 1, 1, 12, 0, 0, tzinfo=timezone.utc) +MASTER_KEY = "sk-master-key-for-tests" +KEYS = session_keys_from_master_key(MASTER_KEY) +PRINCIPAL = SessionPrincipal(user_id="user-123", client_id="llm_client_abc") + + +def _access_token() -> str: + minted = mint_session_token(PRINCIPAL, KEYS, NOW) + assert isinstance(minted, MintedSessionToken) + return minted.token.get_secret_value() + + +def _refresh_token() -> str: + minted = mint_session_refresh_token(PRINCIPAL, KEYS, NOW) + assert isinstance(minted, MintedSessionToken) + return minted.token.get_secret_value() + + +def test_kdf_is_deterministic_and_key_length_is_256_bit(): + again = session_keys_from_master_key(MASTER_KEY) + assert again.signing_key.get_secret_value() == KEYS.signing_key.get_secret_value() + assert len(bytes.fromhex(KEYS.signing_key.get_secret_value())) == 32 + + +def test_kdf_domain_separated_from_envelope_keys(): + envelope_keys = envelope_keys_from_master_key(MASTER_KEY) + session_signing = KEYS.signing_key.get_secret_value() + assert session_signing != envelope_keys.signing_key.get_secret_value() + assert session_signing != envelope_keys.encryption_key.get_secret_value() + + +def test_kdf_differs_across_master_keys(): + other = session_keys_from_master_key("sk-a-different-master-key") + assert other.signing_key.get_secret_value() != KEYS.signing_key.get_secret_value() + + +@pytest.mark.parametrize( + "value,expected", + [ + ("Bearer sk-1234", False), + ("sk-1234", False), + ("Bearer llm_env_abc", False), + ("Bearer llm_refresh_abc", False), + ("llm_session_abc", True), + ("Bearer llm_session_abc", True), + ("bearer llm_srefresh_abc", True), + ], +) +def test_is_session_bearer_shaped(value, expected): + assert is_session_bearer_shaped(value) is expected + + +def test_resolve_admits_valid_access_token_with_and_without_scheme(): + token = _access_token() + for value in (token, f"Bearer {token}", f"bearer {token}"): + result = resolve_session_bearer(value, KEYS, NOW) + assert isinstance(result, SessionBearerAdmitted) + assert result.principal == PRINCIPAL + + +def test_resolve_passes_non_session_bearers_through(): + for value in ("Bearer sk-1234", "Bearer llm_env_whatever", "Bearer eyJhbGciOi"): + assert isinstance(resolve_session_bearer(value, KEYS, NOW), NotSessionBearer) + + +def test_resolve_fails_expired_token_closed_and_flags_expiry(): + token = _access_token() + later = NOW + timedelta(seconds=SESSION_TTL_SECONDS + 1) + result = resolve_session_bearer(f"Bearer {token}", KEYS, later) + assert isinstance(result, SessionBearerInvalid) + assert result.expired is True + + +def test_resolve_fails_tampered_token_closed_without_expiry_flag(): + token = _access_token() + tampered = token[:-2] + ("aa" if not token.endswith("aa") else "bb") + result = resolve_session_bearer(f"Bearer {tampered}", KEYS, NOW) + assert isinstance(result, SessionBearerInvalid) + assert result.expired is False + + +def test_resolve_rejects_refresh_token_at_the_edge(): + result = resolve_session_bearer(f"Bearer {_refresh_token()}", KEYS, NOW) + assert isinstance(result, SessionBearerInvalid) + assert result.expired is False + + +def test_resolve_wrong_master_key_fails_closed(): + other_keys = session_keys_from_master_key("sk-rotated-master-key") + result = resolve_session_bearer(f"Bearer {_access_token()}", other_keys, NOW) + assert isinstance(result, SessionBearerInvalid) + + +def test_refresh_grant_opens_for_the_issued_client(): + result = open_session_refresh_bearer(_refresh_token(), KEYS, NOW, expected_client_id="llm_client_abc") + assert isinstance(result, SessionRefreshOpened) + assert result.principal == PRINCIPAL + + +def test_refresh_grant_rejects_a_different_client(): + result = open_session_refresh_bearer(_refresh_token(), KEYS, NOW, expected_client_id="llm_client_other") + assert isinstance(result, SessionRefreshInvalid) + + +def test_refresh_grant_rejects_access_token_presented_as_refresh(): + result = open_session_refresh_bearer(_access_token(), KEYS, NOW, expected_client_id="llm_client_abc") + assert isinstance(result, SessionRefreshInvalid) diff --git a/tests/test_litellm/proxy/_experimental/mcp_server/outbound_credentials/test_session_token.py b/tests/test_litellm/proxy/_experimental/mcp_server/outbound_credentials/test_session_token.py new file mode 100644 index 00000000000..551270f8d4b --- /dev/null +++ b/tests/test_litellm/proxy/_experimental/mcp_server/outbound_credentials/test_session_token.py @@ -0,0 +1,206 @@ +"""Tests for the identity-only gateway session token (mint/open, hostile-input totality).""" + +from datetime import datetime, timedelta, timezone + +import jwt +import pytest +from pydantic import SecretStr, ValidationError + +from litellm.proxy._experimental.mcp_server.outbound_credentials.session_token import ( + MAX_SESSION_TOKEN_BYTES, + SESSION_ISSUER, + SESSION_REFRESH_PREFIX, + SESSION_REFRESH_TTL_SECONDS, + SESSION_TOKEN_PREFIX, + SESSION_TTL_SECONDS, + MintedSessionToken, + NotASessionToken, + OpenedSessionToken, + SessionBadSignature, + SessionExpired, + SessionKeys, + SessionMalformed, + SessionPrincipal, + SessionTokenTooLarge, + is_session_refresh_token, + is_session_token, + mint_session_refresh_token, + mint_session_token, + open_session_refresh_token, + open_session_token, +) + +NOW = datetime(2026, 1, 1, 12, 0, 0, tzinfo=timezone.utc) +KEYS = SessionKeys(signing_key=SecretStr("k" * 32)) +OTHER_KEYS = SessionKeys(signing_key=SecretStr("x" * 32)) +PRINCIPAL = SessionPrincipal(user_id="user-123", client_id="llm_client_abc") + + +def _mint_access() -> str: + minted = mint_session_token(PRINCIPAL, KEYS, NOW) + assert isinstance(minted, MintedSessionToken) + return minted.token.get_secret_value() + + +def _mint_refresh() -> str: + minted = mint_session_refresh_token(PRINCIPAL, KEYS, NOW) + assert isinstance(minted, MintedSessionToken) + return minted.token.get_secret_value() + + +def _sign_claims(payload: dict, prefix: str = SESSION_TOKEN_PREFIX, keys: SessionKeys = KEYS) -> str: + return prefix + jwt.encode(payload, keys.signing_key.get_secret_value(), algorithm="HS256") + + +def _valid_claims(**overrides) -> dict: + base = { + "iss": SESSION_ISSUER, + "iat": int(NOW.timestamp()), + "exp": int((NOW + timedelta(seconds=600)).timestamp()), + "kind": "session", + "user_id": "user-123", + "client_id": "llm_client_abc", + } + return {**base, **overrides} + + +def test_access_round_trip_recovers_principal_and_caps_ttl(): + minted = mint_session_token(PRINCIPAL, KEYS, NOW) + assert isinstance(minted, MintedSessionToken) + assert minted.expires_at == NOW + timedelta(seconds=SESSION_TTL_SECONDS) + token = minted.token.get_secret_value() + assert is_session_token(token) + assert not is_session_refresh_token(token) + opened = open_session_token(token, KEYS, NOW) + assert isinstance(opened, OpenedSessionToken) + assert opened.principal == PRINCIPAL + + +def test_refresh_round_trip_recovers_principal_and_caps_ttl(): + minted = mint_session_refresh_token(PRINCIPAL, KEYS, NOW) + assert isinstance(minted, MintedSessionToken) + assert minted.expires_at == NOW + timedelta(seconds=SESSION_REFRESH_TTL_SECONDS) + token = minted.token.get_secret_value() + assert is_session_refresh_token(token) + opened = open_session_refresh_token(token, KEYS, NOW) + assert isinstance(opened, OpenedSessionToken) + assert opened.principal == PRINCIPAL + + +def test_access_token_reprefixed_as_refresh_is_rejected_by_signed_kind(): + body = _mint_access().removeprefix(SESSION_TOKEN_PREFIX) + swapped = SESSION_REFRESH_PREFIX + body + assert isinstance(open_session_refresh_token(swapped, KEYS, NOW), SessionMalformed) + + +def test_refresh_token_reprefixed_as_access_is_rejected_by_signed_kind(): + body = _mint_refresh().removeprefix(SESSION_REFRESH_PREFIX) + swapped = SESSION_TOKEN_PREFIX + body + assert isinstance(open_session_token(swapped, KEYS, NOW), SessionMalformed) + + +def test_refresh_token_is_not_an_access_token_at_the_edge(): + assert isinstance(open_session_token(_mint_refresh(), KEYS, NOW), NotASessionToken) + + +def test_expired_access_token_is_expired_not_malformed(): + token = _mint_access() + at_expiry = NOW + timedelta(seconds=SESSION_TTL_SECONDS) + assert isinstance(open_session_token(token, KEYS, at_expiry), SessionExpired) + after = NOW + timedelta(seconds=SESSION_TTL_SECONDS + 1) + assert isinstance(open_session_token(token, KEYS, after), SessionExpired) + + +def test_still_valid_one_second_before_expiry(): + token = _mint_access() + just_before = NOW + timedelta(seconds=SESSION_TTL_SECONDS - 1) + assert isinstance(open_session_token(token, KEYS, just_before), OpenedSessionToken) + + +def test_tampered_signature_is_bad_signature(): + token = _mint_access() + tampered = token[:-2] + ("aa" if not token.endswith("aa") else "bb") + assert isinstance(open_session_token(tampered, KEYS, NOW), SessionBadSignature) + + +def test_key_rotation_invalidates_outstanding_tokens(): + token = _mint_access() + assert isinstance(open_session_token(token, OTHER_KEYS, NOW), SessionBadSignature) + + +@pytest.mark.parametrize( + "candidate,expected", + [ + ("sk-1234", NotASessionToken), + ("llm_env_something", NotASessionToken), + ("", NotASessionToken), + (SESSION_TOKEN_PREFIX, SessionMalformed), + (SESSION_TOKEN_PREFIX + "not-a-jwt", SessionMalformed), + (SESSION_TOKEN_PREFIX + "\ud800garbage", SessionMalformed), + (SESSION_TOKEN_PREFIX + "a" * (MAX_SESSION_TOKEN_BYTES + 1), SessionMalformed), + ], +) +def test_hostile_candidates_never_raise(candidate, expected): + assert isinstance(open_session_token(candidate, KEYS, NOW), expected) + + +def test_multibyte_candidate_over_byte_cap_but_under_char_cap_is_rejected(): + filler = "€" * (MAX_SESSION_TOKEN_BYTES // 3) + candidate = SESSION_TOKEN_PREFIX + filler + assert len(candidate) <= MAX_SESSION_TOKEN_BYTES + assert isinstance(open_session_token(candidate, KEYS, NOW), SessionMalformed) + + +def test_alg_none_token_is_rejected(): + unsigned = jwt.api_jws.encode(b'{"iss":"litellm-mcp-gateway"}', key=None, algorithm="none") + assert isinstance(open_session_token(SESSION_TOKEN_PREFIX + unsigned, KEYS, NOW), SessionMalformed) + + +@pytest.mark.parametrize( + "claims", + [ + _valid_claims(iss="wrong-issuer"), + _valid_claims(exp=str(int((NOW + timedelta(seconds=600)).timestamp()))), + _valid_claims(iat="evil"), + _valid_claims(kind="access"), + _valid_claims(user_id=""), + _valid_claims(nbf=0), + {k: v for k, v in _valid_claims().items() if k != "client_id"}, + {k: v for k, v in _valid_claims().items() if k != "exp"}, + ], +) +def test_signed_but_malformed_claims_are_rejected_without_raising(claims): + token = _sign_claims(claims) + assert isinstance(open_session_token(token, KEYS, NOW), SessionMalformed) + + +def test_signed_claims_with_exact_shape_open(): + token = _sign_claims(_valid_claims()) + opened = open_session_token(token, KEYS, NOW) + assert isinstance(opened, OpenedSessionToken) + assert opened.principal.user_id == "user-123" + + +def test_oversized_client_id_fails_mint_with_typed_error_not_truncation(): + principal = SessionPrincipal(user_id="user-123", client_id="c" * (MAX_SESSION_TOKEN_BYTES + 100)) + minted = mint_session_token(principal, KEYS, NOW) + assert isinstance(minted, SessionTokenTooLarge) + assert minted.max_bytes == MAX_SESSION_TOKEN_BYTES + + +def test_empty_principal_fields_rejected_at_construction(): + with pytest.raises(ValidationError): + SessionPrincipal(user_id="", client_id="c") + with pytest.raises(ValidationError): + SessionPrincipal(user_id="u", client_id="") + + +def test_short_signing_key_rejected_at_construction(): + with pytest.raises(ValidationError): + SessionKeys(signing_key=SecretStr("short")) + + +def test_minted_token_repr_never_leaks_value(): + minted = mint_session_token(PRINCIPAL, KEYS, NOW) + assert isinstance(minted, MintedSessionToken) + assert minted.token.get_secret_value() not in repr(minted)