Skip to main content
Glama
Dney Guardrails MCP Server

Dney Guardrails MCP Server

dney-mcp

Dney Guardrails MCP server. Exposes Dney's guardrail engine as MCP tools for Claude Desktop, Claude.ai, and Claude Code — check content for policy violations, PII, toxicity, and more.

Tools

  • check_content(text, direction) — check a message against active guardrails.

  • check_input_output(input_text, output_text) — check a full exchange.

  • list_guardrails() — list configured guardrails.

  • guardrail_stats() — activity summary.

  • recent_activity(limit) — recent guardrail decisions (metadata only).

  • report_decision(audit_id, verdict) — report that a decision was wrong (false_block or miss). Carries no text: only the decision's id and the verdict.

Related MCP server: Nura MCP Policy Interceptor

Auth

The server resolves the caller's Dney org in one of three ways. Prefer the first two — they keep the secret out of URLs and logs:

  1. Header key (recommended) — x-dney-key (or legacy x-api-key), or the DNEY_API_KEY env var for single-user/local use.

  2. Auth0 OAuth (recommended, optional) — a keyless /mcp request with Authorization: Bearer <Auth0 access token>. The server verifies the token (RS256, issuer, audience, expiry), requires email_verified: true, extracts the user's email from the namespaced AUTH0_EMAIL_CLAIM, and resolves the org by email via the backend's internal service endpoint using MCP_SERVICE_TOKEN. If no Dney account is linked to that email, the tool returns a clear "link your Dney account" error.

  3. Path key (deprecated) — https://mcp.dney.dev/<your-asg-key>/mcp. The PathKeyASGIApp wrapper strips the leading key segment, injects it as x-dney-key, and rewrites the path to /mcp. This exists only for connector UIs that accept a URL and no custom headers.

    ⚠️ Security risk — avoid when possible. A key in the URL path is written to ALB/CloudFront/proxy access logs, APM traces, and browser history, i.e. the key leaks into logs. Use OAuth or a header key wherever the client allows it. Treat path keys as low-privilege, rotate them readily, and never paste the URL into referrer-leaking contexts.

OAuth is enforced only when both AUTH0_ISSUER and MCP_AUDIENCE are set. When either is unset the server runs in legacy path-key-only mode. When OAuth is enabled, a keyless /mcp request without a valid Bearer token receives a 401 with a WWW-Authenticate: Bearer resource_metadata="…" challenge pointing at GET /.well-known/oauth-protected-resource (RFC 9728 protected-resource metadata).

Any client-supplied x-dney-email header is always stripped at the ASGI boundary; only an email verified from a Bearer JWT is ever trusted.

Environment variables

Var

Default

Purpose

DNEY_API_URL

http://localhost:8000

Dney backend base URL

DNEY_API_KEY

—

Single-user/local API key

MCP_HOST / MCP_PORT

0.0.0.0 / 3001

Bind address

AUTH0_ISSUER

—

Auth0 issuer (trailing slash), e.g. https://YOUR-TENANT.us.auth0.com/

MCP_AUDIENCE

—

Expected token audience, e.g. https://mcp.dney.dev

AUTH0_EMAIL_CLAIM

https://dney.dev/email

Namespaced claim holding the verified user email (bare email is not used unless AUTH0_ALLOW_BARE_EMAIL is set)

AUTH0_EMAIL_VERIFIED_CLAIM

https://dney.dev/email_verified

Namespaced claim asserting the email is verified; must be true. Falls back to the bare email_verified claim if the namespaced one is absent. (JWKS mode only.)

AUTH0_USERINFO_VALIDATION

false

When true, validate the bearer token by calling Auth0 /userinfo (reads standard email/email_verified) instead of JWKS + custom audience. Required for Dynamic Client Registration — DCR clients are third-party apps and can't obtain access tokens for a custom API audience, so remove the tenant Default Audience and enable this.

AUTH0_ALLOW_BARE_EMAIL

false

⚠️ If true, fall back to the standard email claim when the namespaced claim is absent. Widens the impersonation surface — leave off unless you fully control every identity source in the Auth0 tenant.

MCP_PUBLIC_URL

https://mcp.dney.dev

Public base URL advertised in the 401 challenge

MCP_SERVICE_TOKEN

—

Shared secret for the backend resolve-org service call

The OAuth email is trusted only when the token asserts email_verified: true; tokens with a missing/false email_verified are rejected.

AUTH0_ISSUER + MCP_AUDIENCE together enable OAuth; MCP_SERVICE_TOKEN must match the backend's MCP_SERVICE_SECRET.

Development

pip install -e ".[dev]"
pytest

Run

# ASGI app (FastMCP + path/header/OAuth auth + /health)
python -m asgard_mcp.server
# or via uvicorn: uvicorn asgard_mcp.server:app

Contributing and security

See CONTRIBUTING.md. Report vulnerabilities privately as described in SECURITY.md, never in a public issue.

License

Apache License 2.0. See LICENSE.

Tool Schema Changelog

Recent tool additions, removals, and schema changes observed during successful MCP inspections.

Loading tool schema changelog…

    Related MCP Connectors

    Related MCP Servers

    • A
      license
      Not graded
      quality
      D
      maintenance
      MCP server for AI agent security guardrails. Provides input validation, prompt injection detection, PII redaction, output filtering, policy enforcement, rate limiting, and comprehensive audit logging.
      72 npm
      1
      MIT
    • A
      license
      A
      quality
      C
      maintenance
      Enables policy-governed MCP interactions with deterministic authorization, tenant isolation, minimized PII exposure, and human approval gates for sensitive mutations, while producing structured audit events.
      3
      MIT
    • A
      license
      Not graded
      quality
      B
      maintenance
      Enables AI agents and MCP clients to safely access upstream MCP servers through centralized policy enforcement, including allow/deny/approval decisions, schema pinning, circuit breakers, and audit logging.
      MIT