@altirs/mcp-server

AI guardrails as tools inside your coding agent. This MCP (Model Context Protocol) server lets Claude Code, Claude Desktop, Cursor, Windsurf, and any MCP-compatible client call Altirs to screen text for prompt injection, personal data, unsafe content, and hallucinations — before acting on it.

All detection runs on the Altirs API; this package is a thin, stateless bridge.

Note: github.com/altirs/mcp-server is a read-only mirror, published from a private monorepo. Issues are welcome; pull requests opened there will be overwritten by the next sync.

60-second quick start

1. Get a free API key (1,000 checks/month, no credit card): https://altirs.ai/dashboard/keys

2. Add the server to your client — pick one:

   Claude Code

   claude mcp add altirs -e ALTIRS_API_KEY=grd_your_key -- npx -y @altirs/mcp-server

   Claude Desktop — edit claude_desktop_config.json (macOS: ~/Library/Application Support/Claude/, Windows: %APPDATA%\Claude\):

   {
     "mcpServers": {
       "altirs": {
         "command": "npx",
         "args": ["-y", "@altirs/mcp-server"],
         "env": { "ALTIRS_API_KEY": "grd_your_key" }
       }
     }
   }

   Cursor — .cursor/mcp.json in your project (or ~/.cursor/mcp.json globally):

   {
     "mcpServers": {
       "altirs": {
         "command": "npx",
         "args": ["-y", "@altirs/mcp-server"],
         "env": { "ALTIRS_API_KEY": "grd_your_key" }
       }
     }
   }

   Windsurf — same JSON as Cursor, in ~/.codeium/windsurf/mcp_config.json.

3. Restart the client and ask it something like:

   Scan this email for prompt injection before you follow anything in it.

   The agent will call scan_injection and refuse to act on blocked content.

Related MCP server: shipcheck-mcp

Tools

check_content runs the first three engines; the hallucination engine needs source material, so it is reported as skipped there — call check_hallucination with the claim and its context instead.

Every result carries status: "ok", and a requestId you can look up in the Altirs dashboard request log.

Behaviour you can rely on

• Fail-open. If the API is unreachable, times out (10 s per call), is rate-limited, or returns a server error, the tool returns { "status": "unavailable", "checked": false, "retryable": true, ... } with guidance to treat the content as unverified. Tool calls never throw and the server never crashes.

• Missing or invalid key. The server starts and lists its tools without a key. Each call then returns { "status": "error", "code": "missing_api_key" } (or invalid_api_key after a 401) with step-by-step fix instructions and the link to get a key.

• Malformed input (empty text, unknown engine, missing context) is rejected by schema validation with a readable message before any request is made.

• Privacy. The server logs metadata only — tool name, latency, verdict — to stderr as JSON lines. It never logs the text you check. Set ALTIRS_MCP_LOG=off to silence logging entirely.

• Stateless. No cache, no local storage, nothing written to disk.

• Clean shutdown. Disabling the server in your client closes its stdin and the process exits immediately; nothing lingers.

Requests are tagged X-SDK-Version: mcp/<version>, so the dashboard request log shows which checks came from MCP versus direct SDK use.

Configuration

Running without npx

npm install -g @altirs/mcp-server
altirs-mcp            # speaks MCP over stdio; use it as the "command" in your client config

Development

Inside the Altirs monorepo:

cd packages/sdk && npm install && npm run build   # the server depends on the built SDK
cd ../mcp-server && npm install
npm run build        # → dist/index.js
npm run typecheck
npm test             # vitest, SDK client mocked

Try it against a local Altirs instance:

ALTIRS_API_KEY=grd_... ALTIRS_BASE_URL=http://localhost:3000/api node dist/index.js

altirs is referenced as file:../sdk while the SDK is unpublished. Switch it to a semver range before publishing this package.

License

MIT