EVM vanity address difficulty, non-custodial delegated grinding, and EIP-712 provenance attestations.
Five tools with complete schemas, strong descriptions (avg 180 chars), and proper parameter typing via Zod. Tool names follow verb_noun pattern (vanity_quote, vanity_grind, etc.). All parameters have descriptions and type constraints. Output schemas are documented in code comments. Tool annotations present (readOnlyHint, destructiveHint, idempotentHint). Main gaps: no pagination support for list-like operations, error messages could be more actionable for LLM recovery, and attestation tool lacks explicit output schema documentation.
Grind a secp256k1 keypair whose address matches a pattern, using every core of the machine running this MCP server. The base scalar is a full 256-bit CSPRNG draw, never a narrowed seed, and every match is re-derived and checked before it is returned. The key does cross the MCP transport into the calling assistant, so treat the result as sensitive and prefer vanity_split_key_grind for anything valuable.
Validate an EVM address, check whether its casing satisfies the EIP-55 checksum (a mixed-case address that fails it is a typo, not an address), read the vanity pattern it appears to carry, and score how much grinding work that represents. Pure maths, no network.
Difficulty for an EVM vanity pattern: probability, expected attempts, p50/p90/p99, rarity tier, and the EIP-55 case multiplier. Every nibble of an EVM address is uniform, so a prefix costs the same as a suffix; the only free variable is casing, where a mixed-case pattern costs 2x per letter. Pure maths, no network.
Delegate a hard vanity pattern to a remote grinder using split-key arithmetic. A secret scalar k1 is generated here and never sent; the remote service searches offsets k2 against the public point P1 = k1*G and returns k2, which is useless without k1. The final key is combined and verified locally, and on secp256k1 it is an ordinary private key that imports anywhere.
vanity_verify_attestation lacks explicit output schema documentation. The handler returns verification results but the response structure is not documented in code or descriptions, forcing LLMs to infer the shape.
Error messages in vanity_grind timeout path are descriptive but could include recovery guidance. 'no match within Xms' tells the LLM the problem but not which tool to try next (vanity_split_key_grind is mentioned but not as a direct next step).
vanity_split_key_grind and vanity_grind both accept prefix/suffix/caseSensitive but lack mutual exclusivity documentation. An LLM might pass both tools when one would suffice, wasting compute.
No batch or multi-pattern variant. If an LLM wants to price 10 patterns, it must call vanity_quote 10 times sequentially. A batch_vanity_quote accepting an array would reduce token waste and latency.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | C | 64 | 2026-07-28+ | v2 |
Verify an EIP-712 vanity attestation: the pattern, the difficulty claim, the freshness nonce, the split-key non-custody equation, and the recovered signer. Every claim is recomputed rather than trusted. Fetches the issuer list when reachable, and says so when it is not.
vanity_inspect_address returns rarity and difficulty but does not document whether these are always present or conditional. Output schema clarity would help LLMs plan downstream logic.