SCML MCP Server
This server does not currently have a deployable release on Glama. Follow the repository's setup instructions to install and configure it for your MCP client, including any required credentials and local dependencies.
SCML MCP Server
Security middleware for LLM agents. Authorize tool calls, scan for injection, quarantine poisoned memory, and redact PII — all through MCP.
Prerequisites
A running SCML mediator (see below).
Policies configured for the agents that will use this server. The MCP server is only a proxy: it forwards your tool-call requests to the mediator, and the mediator decides based on the
agent_idyou pass. There is no automatic policy discovery.
Policy model (read this first). SCML authorizes per agent. Every authorize_tool_call request carries an agent_id; the mediator looks up policy.agents[<agent_id>] and, if the agent is unknown, falls back to the default agent, which is deny-all. Concretely:
On your mediator, write a policy for your agent — e.g.
freight_bookingwithallowed_tools: [book_shipment, get_rate, create_waybill]. Seepolicies/in the SCML repo.Your AI calls
authorize_tool_call(..., agent_id="freight_booking").The mediator allows only tools on that allow-list; anything else (and any unknown
agent_id) is denied.
Two consequences:
If you omit
agent_id, it defaults to"default"and everything is denied. Always pass the agent id you configured policy for.The MCP server relays one mediator's policy space. For multi-tenant setups (e.g. several freight customers), give each customer their own mediator instance and point this server at it via
SCML_URL+SCML_API_KEY. There is no per-user policy scoping at the MCP layer.
Start a local mediator for development:
pip install "trust-mediator[server]"
DATABASE_URL="" REDIS_URL="" TRUST_MEDIATOR_API_KEYS="" \
uvicorn trust_mediator.api.app:app --port 8000A running SCML mediator:
pip install "trust-mediator[server]"
DATABASE_URL="" REDIS_URL="" TRUST_MEDIATOR_API_KEYS="" \
uvicorn trust_mediator.api.app:app --port 8000Related MCP server: Agent Guardrail MCP
Install
pip install -r requirements.txtRun
SCML_URL=http://localhost:8000 SCML_API_KEY=sk-your-key python server.pyClaude Desktop config
Add to ~/.claude/claude_desktop_config.json:
{
"mcpServers": {
"scml": {
"command": "python",
"args": ["/path/to/server.py"],
"env": {
"SCML_URL": "http://localhost:8000",
"SCML_API_KEY": ""
}
}
}
}Tools
Tool | What it does |
| Check if a tool call is allowed by policy — pass your configured |
| Detect prompt injection in external content |
| Score a memory write for integrity |
| Strip PII and secrets from responses |
| Replay session decisions with hash chain |
| Show active security policy |
| Verify SCML mediator is running |
Example: freight company policy
# policies/freight.yaml — load via PUT /v1/policy on your mediator
version: 1
agents:
default:
allowed_tools: [] # deny-all fallback for unconfigured agents
freight_booking:
allowed_tools:
- book_shipment
- get_rate
- create_waybill
- void_shipment
rate_limits:
tool_calls_per_minute: 60Tell your agent to pass agent_id="freight_booking", and SCML will deny any tool call outside that allow-list.
License
Apache 2.0
Related MCP Connectors
MCP gateway with runtime security policy, tool-call-level control, and audit of agent actions.
Scan prompts for injection attacks, redact PII, and audit LLM SDK usage from any MCP client
MCP gateway: one connection for your AI apps, with tool policies, PII redaction and audit logs.
Security firewall for AI agents — scans MCP calls for injection, secrets, and risks.
Related MCP Servers
- AlicenseNot gradedqualityDmaintenanceEnables secure interaction between LLMs and MCP tools by applying zero-trust security controls, including sensitive data masking, file system protection, and policy enforcement.MIT
- AlicenseAqualityDmaintenanceProvides prompt injection detection, PII/secrets redaction, and an audit trail for AI agents via MCP tools.4MIT
- AlicenseNot gradedqualityBmaintenanceSecurity middleware for LLM agents. Authorize tool calls, scan for injection, quarantine poisoned memory, and redact PII — all through MCP. 0% injection success rate on 1,054 tested attacks.Apache 2.0
- AlicenseNot gradedqualityFmaintenanceEnables transparent security for any MCP server by intercepting tool calls, blocking prompt injection attempts, masking PII in responses, and writing immutable audit logs.2,302 npmMIT