Mcp Guard

sidclaw-mcp-guard

Stop AI agents from doing dangerous things through MCP.

Wraps any MCP server. Allows safe calls. Blocks dangerous ones. Holds the rest for you.

---

Get started

See it in action (30 seconds)

npx sidclaw-mcp-guard@latest demo

Set up a real guarded MCP server (2 minutes)

npx sidclaw-mcp-guard@latest quickstart

Creates a config, writes .mcp.json for Claude Code, and starts the approval dashboard.

---

Policy rules — no regex needed

Rules use semantic patterns — human-readable shortcuts instead of raw regex:

# sidclaw.config.yaml

rules:
  - name: allow-reads
    description: Read-only queries are safe
    match:
      pattern: sql-read           # SELECT, EXPLAIN, SHOW
    action: allow

  - name: approve-writes
    description: Data changes need approval
    match:
      pattern: sql-write          # INSERT, UPDATE, DELETE
    action: approve

  - name: deny-destructive
    description: Schema changes are never allowed
    match:
      pattern: sql-destructive    # DROP, TRUNCATE, ALTER, CREATE
    action: deny

default: deny

Available patterns: sql-read, sql-write, sql-destructive, file-read, file-write, file-delete, shell-safe, shell-risky, shell-destructive.

> Note: In config, the action is called approve (what the user writes). In output, it displays as HOLD (describing the state of the call while waiting for human review).

Power users can still use regex via match.args — see [docs/config.md](docs/config.md).

---

Approve from your browser

Every guard instance can run a local approval dashboard:

sidclaw-mcp-guard --ui --upstream npx --upstream-args "..."

Open http://localhost:9091 — see pending requests, approve or deny with one click, inspect the audit trail.

Or use the CLI: npx sidclaw-mcp-guard approve

---

Plain-English explanations

Every decision explains itself:

✔ ALLOW   SELECT * FROM users
  Allowed: read query on users. Read-only queries are safe.

⏳ HOLD    DELETE FROM users WHERE id = 42
  Held for approval: delete from users. Data changes need approval.

✘ BLOCK   DROP TABLE users
  Blocked: drop users. Schema changes are never allowed.

Explanations appear in the terminal, dashboard, and audit log.

---

Observe mode

Test your policies without blocking anything:

sidclaw-mcp-guard --observe --upstream npx --upstream-args "..."

The guard evaluates every call and logs what it would do, but forwards all calls regardless. Switch to enforce mode when ready.

---

Audit trail

Every decision is logged to .sidclaw/audit.jsonl:

{"timestamp":"...","tool":"query","args":{"sql":"SELECT * FROM users"},"decision":"allow","rule":"allow-reads","explanation":"Allowed: read query on users. Read-only queries are safe."}
{"timestamp":"...","tool":"query","args":{"sql":"DELETE FROM users WHERE id=42"},"decision":"approve","rule":"approve-writes","status":"approved","explanation":"Held for approval: delete from users. Data changes need approval."}
{"timestamp":"...","tool":"query","args":{"sql":"DROP TABLE users"},"decision":"deny","rule":"deny-destructive","explanation":"Blocked: drop users. Schema changes are never allowed."}

---

Works with any MCP server

| Server | What you're guarding | Example config | |--------|---------------------|----------------| | @modelcontextprotocol/server-postgres | SQL queries | [examples/sql-demo](examples/sql-demo/) | | @modelcontextprotocol/server-filesystem | File operations | [examples/filesystem-demo](examples/filesystem-demo/) | | Shell-execution servers | Shell commands | [examples/shell-demo](examples/shell-demo/) | | @modelcontextprotocol/server-github | Repo operations | | | Any custom MCP server | Any tool calls | |

---

CLI

# Get started
sidclaw-mcp-guard quickstart                   Set up a real guarded MCP server
sidclaw-mcp-guard demo                         Quick policy showcase
sidclaw-mcp-guard demo -i                      Interactive — try your own SQL

# Run
sidclaw-mcp-guard --upstream              Start the guard proxy
sidclaw-mcp-guard --ui                         Start proxy + approval dashboard
sidclaw-mcp-guard --observe                    Observe mode (log only)

# Approvals
sidclaw-mcp-guard ui                           Open the approval dashboard
sidclaw-mcp-guard approve                  Approve a pending request
sidclaw-mcp-guard deny                     Deny a pending request
sidclaw-mcp-guard list                         List pending approvals
sidclaw-mcp-guard clean                        Remove stale approval files

---

Full Platform

SidClaw Guard is the local-first entry point to SidClaw. When you need more:

| Need | SidClaw Guard (this) | SidClaw Platform | |------|---------------------|------------------| | Policy rules | YAML with semantic patterns | Visual policy editor | | Approvals | Local dashboard + CLI | Dashboard + Slack + Teams + Telegram | | Audit trail | Local JSONL | Hash-chained, exportable, compliance-ready | | Team workflows | Single user | Multi-reviewer, role-based access | | Integrations | MCP servers | 15+ SDKs (LangChain, Vercel AI, CrewAI...) |

Learn more at sidclaw.com →

---

Docs

• [Quick Start](docs/quickstart.md)
• [SQL Demo](docs/sql-demo.md)
• [Config Reference](docs/config.md)
• [How It Works](docs/how-it-works.md)
• [FAQ](docs/faq.md)

---

Development

git clone https://github.com/sidclawhq/mcp-guard.git
cd mcp-guard
npm install
npm run build
npm test

Run locally:

node dist/cli.js demo
node dist/cli.js quickstart

---

License

Apache 2.0

Source & license

This open-source MCP server is cataloged on AgentStack and links to its original source — we do not rehost the code.

• Author: sidclawhq
• Source: sidclawhq/mcp-guard
• License: Apache-2.0
• Homepage: https://sidclaw.com

Install and usage instructions live in the source repository linked above.