fix(mcp): stop reporting false OAuth success when no token was obtained (#34807)

* docs(code-execution): document HERMES_* env narrowing + passthrough workaround

The execute_code sandbox-child env scrub (108397726, #27303) deliberately
dropped the broad HERMES_ prefix passthrough, keeping only an operational
4-var allowlist (HERMES_HOME/PROFILE/CONFIG/ENV). A script that relied on a
non-secret HERMES_* var (HERMES_BASE_URL, HERMES_KANBAN_DB, HERMES_*_WEBHOOK,
or a plugin-defined one) now sees it unset in the child.

Document the behavior change and the two recovery routes (terminal.env_passthrough
in config.yaml, or required_environment_variables in skill frontmatter), plus
the debug log line that surfaces the drop for diagnosis.

* fix(mcp): stop reporting false OAuth success when no token was obtained

`hermes mcp login` reported "Authenticated — N tool(s) available" for
servers that serve tools/list without auth (e.g. Google's official Drive
MCP server) even when the OAuth flow never completed — dynamic client
registration 400'd because the provider doesn't support RFC 7591, so no
token was ever acquired. Every real tool call then hung until timeout
with no indication of why.

Login now verifies a token actually landed on disk after the probe. When
it didn't, it warns that authentication didn't complete and shows the
config needed to supply a pre-registered client_id/client_secret (the
existing, already-supported workaround for DCR-less providers).

Adds a docs pitfall for Google Drive / Atlassian-style providers.

Fixes #34775
This commit is contained in:
Teknium 2026-05-29 12:32:19 -07:00 committed by GitHub
parent 1cb850b674
commit 27a2c4f36f
No known key found for this signature in database
GPG key ID: B5690EEEBB952194
3 changed files with 115 additions and 0 deletions

View file

@ -205,6 +205,22 @@ def _probe_single_server(
return tools_found
def _oauth_tokens_present(name: str) -> bool:
"""Return True if an OAuth token file exists on disk for ``name``.
Used after ``hermes mcp login`` to distinguish a genuine authentication
from a probe that succeeded only because the server allowed
initialize/tools-list without auth (so no token was ever acquired).
"""
try:
from tools.mcp_oauth import HermesTokenStorage
return HermesTokenStorage(name).has_cached_tokens()
except Exception as exc: # pragma: no cover — defensive
logger.debug("Could not check OAuth tokens for '%s': %s", name, exc)
# Be permissive on unexpected errors: don't block a real success.
return True
def _unwrap_exception_group(exc: BaseException) -> Exception:
"""Extract the root-cause exception from anyio TaskGroup wrappers.
@ -631,6 +647,36 @@ def cmd_mcp_login(args):
# Probe triggers the OAuth flow (browser redirect + callback capture).
try:
tools = _probe_single_server(name, server_config)
# A clean probe is NOT proof of authentication. Some MCP servers
# (notably Google's official Drive server) serve initialize +
# tools/list WITHOUT auth, so the probe lists tools even when the
# OAuth flow never completed — e.g. dynamic client registration
# 400'd because the provider doesn't support RFC 7591. Reporting
# "Authenticated — N tools" in that case is a false success: every
# real tool call later hangs until timeout because there's no token.
# Verify a token actually landed on disk before claiming success.
if not _oauth_tokens_present(name):
_warning(
"Server responded, but no OAuth token was obtained — "
"authentication did not complete."
)
print()
_info(
"Some providers (e.g. Google Drive, Atlassian) do not support "
"automatic client registration. For those you must create an "
"OAuth client yourself and add its credentials to config.yaml:"
)
print()
print(color(f" mcp_servers:", Colors.DIM))
print(color(f" {name}:", Colors.DIM))
print(color(f" url: {url}", Colors.DIM))
print(color(f" auth: oauth", Colors.DIM))
print(color(f" oauth:", Colors.DIM))
print(color(f" client_id: \"<your-oauth-client-id>\"", Colors.DIM))
print(color(f" client_secret: \"<your-oauth-client-secret>\"", Colors.DIM))
print()
_info("Then re-run `hermes mcp login " + name + "`.")
return
if tools:
_success(f"Authenticated — {len(tools)} tool(s) available")
else:

View file

@ -595,3 +595,58 @@ class TestMcpLogin:
out = capsys.readouterr().out
assert "no URL" in out or "not an OAuth" in out
def test_login_false_success_no_token(self, tmp_path, capsys, monkeypatch):
"""Probe lists tools without auth (Google Drive), but no token landed.
The server allows tools/list without auth (DCR 400'd), so the probe
succeeds yet no OAuth token exists. Login must NOT claim success it
should warn and point the user at pre-registered client_id config.
"""
_seed_config(tmp_path, {
"googledrive": {
"url": "https://drivemcp.googleapis.com/mcp/v1",
"auth": "oauth",
},
})
# Probe returns tools even though auth never completed.
monkeypatch.setattr(
"hermes_cli.mcp_config._probe_single_server",
lambda name, cfg: [("search_files", "d"), ("read_file_content", "d")],
)
# No token file is created → _oauth_tokens_present() returns False.
from hermes_cli.mcp_config import cmd_mcp_login
cmd_mcp_login(_make_args(name="googledrive"))
out = capsys.readouterr().out
assert "no OAuth token was obtained" in out
assert "Authenticated" not in out
assert "client_id" in out
def test_login_genuine_success_with_token(self, tmp_path, capsys, monkeypatch):
"""Probe lists tools AND a token exists → report real success."""
_seed_config(tmp_path, {
"realserver": {"url": "https://mcp.example.com/mcp", "auth": "oauth"},
})
token_dir = tmp_path / "mcp-tokens"
# cmd_mcp_login wipes tokens before probing, then the real OAuth flow
# writes a fresh token during the probe. Simulate that: the mocked
# probe drops a token file, mirroring a successful authorization.
def mock_probe(name, cfg):
token_dir.mkdir(exist_ok=True)
(token_dir / "realserver.json").write_text('{"access_token": "x"}')
return [("a", "d"), ("b", "d"), ("c", "d")]
monkeypatch.setattr(
"hermes_cli.mcp_config._probe_single_server", mock_probe
)
from hermes_cli.mcp_config import cmd_mcp_login
cmd_mcp_login(_make_args(name="realserver"))
out = capsys.readouterr().out
assert "Authenticated — 3 tool(s) available" in out
assert "no OAuth token" not in out

View file

@ -229,6 +229,20 @@ On first connect, Hermes prints an authorize URL, opens your browser when possib
See [OAuth over SSH / Remote Hosts](../../guides/oauth-over-ssh.md#mcp-servers) for the full walkthrough, including DCR-less servers (e.g. Slack), pre-registered `client_id`/`client_secret`, scope customization, and re-auth via `hermes mcp login <server>`.
**Pitfall — providers that don't support automatic registration (Google Drive, Atlassian).** Some servers reject the dynamic client registration step (RFC 7591) that bare `auth: oauth` relies on — Google's official Drive server (`https://drivemcp.googleapis.com/mcp/v1`) returns a `400 Bad Request`, so no OAuth client is created and no token is acquired. The symptom is subtle: these servers also serve `tools/list` *without* auth, so `hermes mcp login` can list the tools and look like it worked, but every real tool call later times out. `hermes mcp login` now detects this (it checks that a token actually landed on disk) and tells you to supply your own OAuth client. Create one in the provider's console and add it to config:
```yaml
mcp_servers:
googledrive:
url: "https://drivemcp.googleapis.com/mcp/v1"
auth: oauth
oauth:
client_id: "<your-oauth-client-id>"
client_secret: "<your-oauth-client-secret>"
```
Then run `hermes mcp login googledrive` — with the pre-registered client, Hermes skips registration and runs the normal browser authorization flow.
**Pitfall — config auto-reload race.** When you edit `~/.hermes/config.yaml` from inside a running Hermes session, the CLI auto-reloads MCP connections with a 30s timeout. That's not enough for an interactive OAuth flow. Add the entry, then run `hermes mcp login <server>` from a fresh terminal — it waits the full 5 minutes for you to complete auth.
## Basic configuration reference