Kaidn MCP

Model Context Protocol server for the Kaidn fraud-scoring API.

Investigate fraud in plain English — "why was this signup blocked?", "what else has this device touched?", "what's in the review queue this morning?"

• Evidence, not just a score. Every reason carries the raw numbers behind it, so a model can explain a verdict rather than guess at it.
• Read-only by default. Nothing changes your tenant unless you opt in.
• Quota-guarded. An agent in a loop cannot spend your month in ten minutes.
• Any client. MCP is an open protocol — stdio locally, Streamable HTTP for remote and hosted agents.

Requirements

Node.js 18 or newer, and an API key from your Kaidn dashboard.

Getting started

First, install the Kaidn MCP server with your client. Standard config works in most of the tools:

{
  "mcpServers": {
    "kaidn": {
      "command": "npx",
      "args": ["@kaidn/mcp@latest"],
      "env": { "KAIDN_API_KEY": "your_key" }
    }
  }
}

claude mcp add kaidn --env KAIDN_API_KEY=your_key -- npx @kaidn/mcp@latest

Add the standard config to claude_desktop_config.json, then restart Claude. Settings → Developer → Edit Config opens the file.

Settings → MCP → Add new MCP Server, or add the standard config to .cursor/mcp.json in your project (or ~/.cursor/mcp.json for every project).

code --add-mcp '{"name":"kaidn","command":"npx","args":["@kaidn/mcp@latest"],"env":{"KAIDN_API_KEY":"your_key"}}'

Add the standard config to ~/.codeium/windsurf/mcp_config.json.

Add the standard config to cline_mcp_settings.json via the MCP Servers icon → Configure MCP Servers.

Add to settings.json under context_servers, using the same command, args and env as the standard config.

Any MCP client takes a command, args and an env block. Use the standard config above. If the client can only reach the server over the network rather than spawning a process, see Streamable HTTP.

Configuration

| Option | Environment variable | Default | Purpose | |---|---|---|---| | | KAIDN_API_KEY | required | Your secret key. Environment only — never a flag, never a tool argument. | | | KAIDN_API_URL | https://api.kaidn.io | API base URL | | --allow-writes | KAIDN_MCP_ALLOW_WRITES=1 | off | Register the mutating tools | | | KAIDN_MCP_MAX_QUOTA_CALLS | 100 | Quota ceiling per process | | --http | KAIDN_MCP_TRANSPORT=http | stdio | Serve Streamable HTTP | | --host <addr> | KAIDN_MCP_HOST | 127.0.0.1 | HTTP bind address | | --port <n> | KAIDN_MCP_PORT | 8765 | HTTP port | | | KAIDN_MCP_HTTP_TOKEN | unset | Require Authorization: Bearer on HTTP | | --help | | | Show usage | | --version | | | Show the version |

Precedence: CLI flags override environment variables.

The API key is deliberately env-only. A key passed as a flag leaks into process listings and shell history.

Transports

| Transport | Use it for | Endpoint | |---|---|---| | stdio (default) | local clients that spawn a subprocess | — | | Streamable HTTP | remote agents, containers, anything off-machine | POST /mcp |

HTTP+SSE is deliberately absent: deprecated in the 2025-03-26 spec and sunset in June 2026.

Streamable HTTP

npx @kaidn/mcp@latest --http --port 8765

Stateless — a fresh server per request, nothing shared between callers — so it sits behind a load balancer without surprises. GET /health is unauthenticated so an orchestrator can check liveness without holding the token.

Docker

docker build -t kaidn-mcp .

# stdio — behaves like the npx invocation
docker run -i --rm -e KAIDN_API_KEY=your_key kaidn-mcp

# HTTP — for remote agents
docker run --rm -p 8765:8765 \
  -e KAIDN_API_KEY=your_key \
  -e KAIDN_MCP_TRANSPORT=http \
  -e KAIDN_MCP_HOST=0.0.0.0 \
  -e KAIDN_MCP_HTTP_TOKEN=your_token \
  kaidn-mcp

Multi-stage build, runs as the unprivileged node user, with a healthcheck.

Security

The server holds your API key. Whoever can reach it can spend your quota, so the defaults are conservative and the guards fail closed rather than warning.

• Binds 127.0.0.1, and refuses to start on a wider interface unless KAIDN_MCP_HTTP_TOKEN is set. It stops with an explanation rather than quietly exposing your account.
• Read-only by default. add_to_list and label_outcome exist only with --allow-writes.
• set_config and forget_subject are never exposed, in any mode. One silently changes the verdict on every future event; the other is irreversible GDPR erasure. Both belong in the dashboard, in front of a human.
• Quota ceiling per process, with remaining budget reported on every costing response. A reservation that would overshoot is refused outright rather than partially spent.
• The key never crosses the tool boundary — not as a parameter, not in output, not in an error.

Tools

Two things govern every tool: whether it spends quota, and whether it changes anything.

Read-only — available by default

| Tool | Cost | What it does | |---|---|---| | get_stats | free | Verdict, score and reason rollups over a rolling window. Start here. | | list_events | free | Scored events, newest first, filterable by verdict or type | | explain_event | free | Every check that fired on one event, with the raw evidence | | triage_queue | free | Everything on review, highest score first | | get_config | free | Effective weights and thresholds for this tenant | | investigate_entity | 1 row¹ | Enrichment, network reputation and related events for one entity | | check_email | 1 row | Disposable domain, deliverability, fraud score, abuse history | | check_ip | 1 row | Proxy, VPN, Tor, datacenter ASN, geo, abuse history | | check_phone | 1 row | Validity, line type, carrier, fraud score | | score_event | 1 row | Score a new event (also records it) |

¹ Free when the entity is a device_id; enrichment only costs on email or IP.

Mutating — require --allow-writes

| Tool | What it does | |---|---| | add_to_list | Add an entity to the allow or block list | | label_outcome | Report a confirmed fraud / chargeback / legit outcome |

Worked examples

The tools are designed to be chained. These are the flows they were built for.

Morning triage

You: What happened overnight, and what needs me?

The model calls get_stats for the shape of the last 24 hours, then triage_queue for the events sitting on review, then explain_event on the worst one. You get a ranked list with the reasoning attached, rather than a dashboard you still have to read.

"Why was this customer blocked?"

You: Event evt_8f21c — a customer says they were wrongly blocked.

explain_event returns every check that fired with its raw evidence — the datacenter ASN it matched, how many accounts shared the device, the velocity count. Enough to answer the customer, or to conclude the rule was wrong and needs tuning.

Working outward from one signal

You: Is 194.x.x.x a one-off or part of a ring?

investigate_entity returns enrichment and network reputation for the IP plus every recent event it appears in. If the same device ids keep recurring, that is a ring rather than a coincidence.

Checking a rule change before making it

You: If I dropped the velocity weight, what would stop being blocked?

get_config reads the current weights; list_events with verdict: "block" shows what is currently caught. The model can tell you which of those hang on the check you are about to weaken.

Error handling

Failures come back as tool errors with a readable message, not exceptions — the model can act on them.

| You see | Meaning | Fix | |---|---|---| | KAIDN_API_KEY is not set | Server started without a key | Set it in the client's env block | | Kaidn error: 401 … | Key rejected | Rotate or re-copy it from the dashboard | | Kaidn error: 429 … | Rate limited | Slow down; per-key throttling is by the minute | | Session quota ceiling reached (100/100 …) | The guard stopped an expensive run | Raise KAIDN_MCP_MAX_QUOTA_CALLS deliberately, or restart | | No event <id> in the most recent 200 events | Event is older than the scan window | Page back with list_events using offset | | Supply exactly one of email, ip or device_id | Ambiguous investigation | Ask about one entity at a time | | Refusing to bind <host> without authentication | Non-loopback HTTP with no token | Set KAIDN_MCP_HTTP_TOKEN, or bind 127.0.0.1 |

Errors never contain your API key.

Troubleshooting

The client shows no tools. Check the client's MCP log for the startup line. kaidn-mcp: ready (stdio, …) on stderr means the server is up and the problem is on the client side. Nothing at all usually means npx could not resolve the package or Node is older than 18.

It starts, then exits immediately. Almost always a missing KAIDN_API_KEY. The message says so on stderr; some clients hide stderr, so run it in a terminal to see it.

add_to_list and label_outcome are missing. Working as designed. They need --allow-writes.

set_config and forget_subject are missing. Also by design, and they are not available in any mode. See SECURITY.md.

HTTP mode refuses to start. You bound something other than loopback without a bearer token. That is the guard working — the process holds your API key.

Everything is slow. The enrichment checks make live upstream calls. get_stats, list_events, explain_event and triage_queue are free and fast; prefer them when reading history.

Check the server independently of the client:

node dist/index.js --help                 # no key required
KAIDN_API_KEY=your_key npm start          # should print a ready line

Support

• Bugs and feature requests: GitHub issues
• Security: [email protected] — see SECURITY.md
• Privacy and data handling: PRIVACY.md
• The API itself: kaidn.io

Run from source

git clone https://github.com/Kaidn-io/kaidn-mcp.git
cd kaidn-mcp
npm install
npm run build
npm test

claude mcp add kaidn --env KAIDN_API_KEY=your_key -- node /absolute/path/to/kaidn-mcp/dist/index.js

To check it starts without a client:

KAIDN_API_KEY=your_key npm start

It prints kaidn-mcp: ready (stdio, read-only, quota ceiling 100) to stderr and then waits on stdin — that is the MCP transport, so the silence is correct.

Why the evidence matters

Kaidn's engine is rules-first and explainable: every reason carries the raw numbers behind it. A bare score gives a model nothing to reason about, while checks[] with evidence attached gives it something to explain. That is the difference between explain_event being useful and being decorative.

Rules decide. The model narrates.

Project

• CONTRIBUTING.md — what belongs here, and the guarantees a change must not break
• SECURITY.md — reporting, threat model, known limitations
• PRIVACY.md — what passes through, what is stored, what is not
• CODE_OF_CONDUCT.md

Licence

MIT