langchain-ai/deepagents · error · RuntimeError

No MCP OAuth provider matched {server_url!r}

Error message

No MCP OAuth provider matched {server_url!r}

What it means

resolve_provider scans the internal OAuth provider registry and returns the first provider whose `matches(server_url)` is true; the generic provider is documented to always match. Raising here means the registry has no matching entry — effectively a broken/incomplete registry build rather than a user config problem.

Source

Thrown at libs/code/deepagents_code/mcp_providers/_registry.py:39

def resolve_provider(server_url: str) -> OAuthProvider:
    """Return the provider policy that owns `server_url`.

    Args:
        server_url: Remote MCP endpoint URL.

    Returns:
        The first matching `OAuthProvider`; falls back to `GenericProvider`.

    Raises:
        RuntimeError: If no provider matches (unreachable in practice since
            `GenericProvider.matches` always returns `True`).
    """
    for provider in _REGISTRY:
        if provider.matches(server_url):
            return provider
    msg = f"No MCP OAuth provider matched {server_url!r}"
    raise RuntimeError(msg)

View on GitHub (pinned to a1af029e6e)

Solutions

1. Check whether _REGISTRY was modified/cleared in your session or test setup; restore the default registry including the generic fallback provider.
2. Reinstall deepagents-code to restore the complete built-in registry.
3. If you maintain a custom registry, ensure GenericProvider (whose matches() always returns True) is registered last.
4. Verify the server_url is well-formed; a malformed URL combined with strict provider predicates can skip all specific providers.

Example fix

// before (test setup)
_registry_override[:] = [GitHubProvider]
// after
_registry_override[:] = [GitHubProvider, GenericProvider()]

from deepagents_code.mcp_providers import _registry
assert _registry._REGISTRY, "OAuth provider registry is empty"

try:
    provider = resolve_provider(server_url)
except RuntimeError as exc:
    if str(exc).startswith("No MCP OAuth provider matched"):
        provider = GenericProvider()  # fallback for custom registries
    else:
        raise

When it happens

Trigger: Calling resolve_provider (via build_oauth_provider or login) with a server_url when _REGISTRY is empty or its entries' matches() predicates all fail — i.e. the always-matching GenericProvider is missing from the registry.

Common situations: Test code monkeypatching/clearing _REGISTRY; custom builds that registered only specific providers (e.g. GitHub) and removed the generic fallback; plugin code that filtered the registry before login.

Related errors

• MCPReauthRequiredError({server_name})
• Authorization denied by provider: {err_code}{detail}
• Callback URL is missing the 'code' parameter.
• Device code request failed: HTTP {response.status_code} from
• Device code response from {device_code_url} is missing requi

AI-assisted analysis of langchain-ai/deepagents@a1af029e6e (2026-08-29). Data as JSON: /api/errors/379ddc28aaa5c931. Report an issue: GitHub.