agentscope-ai/agentscope · error · ValueError

MCPClient name ' ' contains characters not allowed by LLM…

Error message

MCPClient name '{self.name}' contains characters not allowed by LLM providers (only [a-zA-Z0-9_-] are permitted). Please rename it.

What it means

MCPClient validates in model_post_init that its name matches ^[a-zA-Z0-9_-]+$ because the name is used to compose model-facing tool names (mcp__{name}__{tool}) that LLM providers restrict to that character set. Any space, dot, slash, unicode, or other symbol in the MCP server name triggers this ValueError at construction time.

Solutions

Rename the MCP client using only letters, digits, underscore, and hyphen (e.g. "github-api" instead of "github.com/api")
If the name comes from a config file, sanitize it before constructing MCPClient
Keep the display/description separate from the internal name if human-friendly naming is needed

Example fix

// before
client = MCPClient(name="my search server", mcp_config=config)

// after
client = MCPClient(name="my-search-server", mcp_config=config)

Defensive patterns

Strategy: validation

Validate before calling

import re
assert re.fullmatch(r"[a-zA-Z0-9_-]+", name), f"invalid MCP name: {name!r}"
client = MCPClient(name=name, mcp_config=cfg)

Try / catch

try:
    MCPClient(name=name, mcp_config=cfg)
except ValueError as e:
    if "not allowed by LLM providers" in str(e):
        name = re.sub(r"[^a-zA-Z0-9_-]", "-", name)

Prevention

Derive MCP names with a sanitizer: re.sub(r'[^a-zA-Z0-9_-]', '-', raw_name)
Keep a naming convention (kebab-case) for MCP servers in config files
Add a config lint step before constructing clients

When it happens

Trigger: MCPClient(name="my mcp.server", mcp_config=...) or any name containing spaces, dots, colons, or non-ASCII characters; also names loaded from config files (YAML/JSON) where naming rules were not enforced.

Common situations: Naming an MCP server after a URL or path (e.g. "github.com/api"), using display names with spaces, copying server names from other tools (Claude Desktop config uses arbitrary names) into agentscope.

Related errors

AI-assisted analysis of agentscope-ai/agentscope@e90f1c7592 (2026-08-28). Data as JSON: /api/errors/b4721ad23c02457a. Report an issue: GitHub.

Appendix: source

Thrown at src/agentscope/mcp/_mcp_client.py:123

    _stack: AsyncExitStack | None = PrivateAttr(default=None)
    _is_connected: bool = PrivateAttr(default=False)
    _cached_tools: list[mcp.types.Tool] | None = PrivateAttr(default=None)

    @property
    def is_connected(self) -> bool:
        """Whether the client is currently connected.

        Returns:
            True if connected, False otherwise.
        """
        return self._is_connected

    def model_post_init(self, __context: Any) -> None:
        """Validate configuration and initialize client."""
        # MCP name is used to compose model-facing tool names
        # (mcp__{name}__{tool}), which must match ^[a-zA-Z0-9_-]+$.
        if not re.fullmatch(r"[a-zA-Z0-9_-]+", self.name):
            raise ValueError(
                f"MCPClient name '{self.name}' contains characters not "
                f"allowed by LLM providers (only [a-zA-Z0-9_-] are "
                f"permitted). Please rename it.",
            )

        # STDIO MCP must be stateful
        if self.mcp_config.type == "stdio_mcp" and not self.is_stateful:
            raise ValueError(
                "STDIO MCP must be stateful (is_stateful=True).",
            )

        # Check arguments for self.enable_tools and disable_tools
        if self.enable_tools is not None:
            if not isinstance(self.enable_tools, list) or any(
                not isinstance(_, str) for _ in self.enable_tools
            ):
                raise ValueError(
                    "Enable tools should be a list of strings, but got "

View on GitHub (pinned to e90f1c7592)