GHSA-rp72-5v5q-2446 json
@cardano402/mcp-server missing spending limits, LAN-exposed HTTP transport, and SSRF via catalog.server.url
published 4mo ago modified 2mo ago
- cvss
- 0.0
- version 3.1, GitHub
- cisa kev
- Not listed
- not in CISA's catalog
- epss
- —
- no score, EPSS covers CVEs only
- public exploits
- 0
- none known
Summary
## Summary `@cardano402/mcp-server` versions `<= 0.1.1` ship three security gaps that can lead to unauthorized fund movement when the package is used as designed (an MCP server exposing Cardano payment tools to an
## Impact ### 1. No spending limits on signed payments An LLM (or prompt-injected LLM) calling tools registered by the MCP server can invoke them in a loop. Each call signs a real Cardano transaction for the catalog-advertised amount. There is no per-call cap, daily ceiling, MCP elicitation/confirmation step, or recipient allowlist. The `MAINNET=true` env-var guardrail can be bypassed by any LLM with shell-tool access. Worst case: full wallet drain.
### 2. HTTP transport binds 0.0.0.0 without authentication `cardano402-mcp --transport http` listens on all interfaces with no `Origin` allowlist, no bearer-token requirement, and no CORS check. Anyone on the same LAN can POST MCP `tools/call` and trigger signed payments from the operator's wallet.
### 3. SSRF via `catalog.server.url` A malicious catalog can declare a `server.url` pointing at internal infrastructure (e.g. `http://169.254.169.254/latest/meta-data`). The `allowInsecure` guard in 0.1.1 only checks the catalog URL itself, not the `server.url` it returns. `endpoint.path` is also not normalized, so `..` traversal or absolute URLs work.
## Patches Fixed in `@cardano402/mcp-server@0.1.2`: - Per-call and per-day spending limits (default 5 ADA / 50 ADA) + optional recipient allowlist + MCP `elicitation/create` confirmation hook. - HTTP transport defaults to `127.0.0.1`; non-loopback requires `--http-bearer-token`; per-request `Origin` allowlist + bearer check. - `catalog.server.url` validated against private-CIDR rules (RFC1918, RFC4193, link-local, CGNAT, multicast, IPv4-mapped IPv6, loopback) unless `CARDANO402_ALLOW_INSECURE=true`. - `endpoint.path` rejected if it contains `..`, NUL, whitespace/CRLF, an absolute URL, or `//host/...`. - Per-tool mainnet opt-in via `--mainnet-confirmed-tools`.
## Workarounds for 0.1.1 users - Do not run with `--transport http` on an untrusted network; use `--transport stdio` (default). - Only point the server at catalogs you control or have audited. - Use a low-balance hot wallet, never your main wallet. - Avoid `MAINNET=true` until upgraded to 0.1.2.
Patch / fix
- npm @cardano402/mcp-server 0.1.2
Is GHSA-rp72-5v5q-2446 in your dependencies? Fencer shows which of your repositories are affected as soon as an advisory lands, and opens the fix as a pull request.
Check with FencerCVSS scores
| assigned by | version | score | vector |
|---|---|---|---|
| GitHub | 3.1 | 0.0 | CVSS:3.1/AV:N/AC:L/PR:N/UI:N/S:U/C:N/I:N/A:N |
Affected packages
| ecosystem | package | affected | fixed in |
|---|---|---|---|
| npm | @cardano402/mcp-server | <0.1.2 |
0.1.2 |
Classification
References
Sources
- GitHub Advisory GHSA-rp72-5v5q-2446
Fencer Intel keeps one entry per vulnerability: the CVE id wins, then GHSA, and records that alias each other are merged, except that two CVEs always keep separate entries. The same data is available as JSON.