# SaferICO

> On-chain security for EVM contracts, tokens and wallets: automated audits, adversarial
> simulation, wallet risk, live contract monitoring, and security testing for AI agents.
> 25 EVM networks. Founded 2021.

This file is generated from the tool registry (scripts/llms.mjs). It describes what an
automated client can actually do here, and what it cannot.

## What SaferICO does

- Audits Solidity with 201 detectors and stores a shareable report.
- Simulates whether the findings chain into a real attack, rather than listing them.
- Analyses UNVERIFIED contracts straight from deployed bytecode.
- Simulates a buy and a sell to detect honeypots and hidden transfer taxes.
- Scores wallet risk and revokes dangerous approvals.
- Watches a live contract for ownership, mint, pause, proxy-upgrade, liquidity and code changes.
- Tests AI agents that hold keys: capability graph, adversarial probes, and a decoy signer that holds no private key.
- Publishes 204 full audit reports and a database of 203 vulnerability classes.

## What it will not do

- It never signs, sends or holds anything. No private key is accepted, stored or requested.
- It never says a contract or a wallet is "safe". It describes risk and shows the evidence.
- It does not run the 201-detector audit on the server. That runs in your browser,
  in your CI runner, or on your machine — so source code does not have to move.

## Interfaces

### MCP (remote) — the lightweight read tools
- Endpoint: https://saferico.com/mcp (POST, JSON-RPC 2.0, Streamable HTTP)
- Protocol revisions: 2026-07-28, 2025-11-25, 2025-06-18, 2025-03-26
- Auth: Authorization: Bearer sfi_live_… — requires the Business ($29.9/mo) plan
- Server card: https://saferico.com/mcp.json

### MCP (local) — the full audit, on your machine
- Command: npx saferico mcp
- Tools: scan_path, scan_source, engine_info
- No API key. No upload. The detectors are downloaded and run locally.

### CLI
- npx saferico scan ./contracts
- npx saferico scan ./contracts --json --output report.json --fail-on high
- npx saferico scan ./contracts --engine-version 2026.09.01   (reproducible)
- Engine manifest: https://saferico.com/engine/manifest.json

### HTTP API
- Base: https://saferico.com/api/v1
- OpenAPI: https://saferico.com/openapi.json
- Auth: Authorization: Bearer sfi_live_…
- Every response has `ok`. On failure, switch on `error`, not on `detail`.

## Tools available remotely

- market_info(): Live native-coin prices and gas across the supported chains, and which chain is cheapest to deploy on right now. [Business ($29.9/mo), 2 credits]
- token_info(token, network?): Read a token's on-chain facts: name, symbol, decimals, total supply and owner. [Business ($29.9/mo), 2 credits]
- check_honeypot(token, network?): Simulate a buy and a sell against a token to detect honeypots and hidden transfer taxes. [Business ($29.9/mo), 2 credits]
- portfolio_scan(address?): Read the wallet's native-coin balance and its USD value across the supported chains. [Business ($29.9/mo), 2 credits]
- contract_source(address, network?): Fetch a contract's verification status, compiler, proxy status and source metadata from the explorer. [Business ($29.9/mo), 2 credits]
- bytecode_scan(address, network?): Analyse an UNVERIFIED contract straight from its deployed bytecode: which functions it exposes (mint, blacklist, pause, fee and withdrawal entry points), which opcodes it contains (SELFDESTRUCT, DELEGATECALL, tx.origin), whether it is an upgradeable proxy, and which compiler built it. Use this whenever contract_source reports the source is not verified. [AI Agents ($59.9/mo), 2 credits]
- explain_transaction(hash, network?): Explain what one transaction actually did: status, decoded function call, ERC-20 and NFT movements, approvals granted (flagging unlimited ones), every contract touched, and security-relevant signals such as proxy upgrades, ownership transfers and role grants. Reads the receipt and its logs — it does NOT trace internal calls, because public RPC endpoints do not expose tracing. [Business ($29.9/mo), 2 credits]
- scan_agent_config(content, filename?): Security-review the text of an AI agent skill, instruction file or MCP server configuration for patterns that are dangerous by construction: shell commands in server definitions, disabled sandboxes, wildcard command permissions, hardcoded credentials and private keys, instructions to read credential stores, outbound calls to data-collection endpoints, prompt-injection and instruction-override text, hidden HTML-comment instructions, invisible or bidirectional-override characters, and unpinned packages. This is pattern matching over text, NOT smart-contract analysis, and a clean result means no known pattern matched — not that the file is safe to run. Reads up to 4 KB here; use `npx saferico skill <path>` for a whole directory, locally, with nothing uploaded. [Business ($29.9/mo), 1 credits]

## Tools that exist but are NOT remote

- audit_contract: Run the full 201-detector security audit on a contract and store the report.
- explain_report: Explain an audit report in plain language.
- simulate_attack: After a full audit, compose its findings into whole attack CHAINS — how an attacker would combine them to actually take funds, with the evidence for each step, the affected functions, the fix, and a comparable real-world incident. Offer this once an audit has produced findings.
- revoke_approvals: Scan a wallet for risky token approvals and revoke them.

These need either a wallet to sign (so they live on the website) or far more CPU than an
edge request may use (so they run locally). Asking for them over MCP or the API returns a
clear refusal, not a partial result.

## Networks

Ethereum, Optimism, BNB Chain, Gnosis, Unichain, Polygon, Monad, Sonic, opBNB, Fraxtal, World Chain, HyperEVM, Sei, Abstract, Mantle, Base, ApeChain, Arbitrum, Celo, Avalanche, Linea, Berachain, Blast, Taiko, Katana

Reading is supported on all 25. Building and paying are supported on fewer —
see https://saferico.com/docs.

## Plans

- Free: $0/mo, 0 credits/mo (key: starter)
- Pro: $9.99/mo, 1500 credits/mo (key: pro)
- Business: $29.9/mo, 4000 credits/mo (key: business)
- AI Agents: $59.9/mo, 12000 credits/mo (key: elite)
- Protocol: $199/mo, 40000 credits/mo (key: protocol)

Anonymous visitors get 3 scans a day; a free account gets 10 scans a month.
The local CLI scan needs no plan at all.

## Rate limits

- 240 requests per minute per address
- 10000 requests per day per key

Both apply to /api/v1 and /mcp and are answered with HTTP 429, `error: "rate-limited"` and a
`Retry-After` header. They bound REQUESTS, not spend: listing the tools and refused calls cost
no credits but still count here. A 429 carrying `credits-exhausted` is a different thing — that
one means the month's credits are gone and retrying will not help.

## Documentation

- Docs: https://saferico.com/docs
- Vulnerability database: https://saferico.com/safe/ (203 classes)
- Published audits: https://saferico.com/audits/
- Hall of hacks: https://saferico.com/hacks/
- Machine-readable card: https://saferico.com/agent.json