MCP Connectors

Every tool call, governed.

Statis ships a first-class MCP adapter. Register your MCP servers, apply policy rules to tool calls, and get a full receipt ledger for every Claude and Cursor interaction.

What is MCP?

The Model Context Protocol is Anthropic's open standard for connecting AI models to external tools. Claude, Cursor, and a growing ecosystem of AI clients use MCP to discover and invoke tools on behalf of users.

Statis sits in front of your MCP servers as a governance layer: every tool call is proposed, evaluated against policy, locked for exactly-once execution, and receipted.

How it works

$ statis connector create \
--name my-mcp-server \
--url https://mcp.example.com \
--auth-type bearer \
--auth-token $MCP_TOKEN
✓ Connector my-mcp-server registered

Policy conditions

Write policy rules that match on MCP server, tool name, and tool arguments. Allow, deny, or escalate any tool call before it runs against the real backend.

rule: mcp_file_write_safety
when:
  action_type: mcp_tool_call
  mcp_server: my-mcp-server
  tool_name: write_file
then:
  condition: path.startswith("/workspace/")
  on_true:  APPROVE
  on_false: ESCALATE

SDK: the mcp_guard decorator

Wrap any MCP handler with the Statis SDK decorator and every tool call flows through the governance layer automatically.

from statis import mcp_guard

@mcp_guard(connector="my-mcp-server")
async def write_file(path: str, content: str):
    # Statis evaluates policy before this runs
    # Receipt written after completion
    return await actually_write_file(path, content)

Supported auth types

  • Bearer token — OAuth2 access tokens, API keys as bearer
  • Basic auth — username/password pairs, base64 encoded automatically
  • API key header — custom header name and value
  • None — public MCP servers

Read the docs

Full reference: docs.statis.dev/mcp →