MCP server exposing koreafilings.com's x402-paid DART summary API as a tool for Claude Desktop, Cursor, Continue, and other MCP clients.
Five tools with clear naming (verb_noun pattern: get_*, find_*, list_*) and comprehensive descriptions (100-250 chars each). All tools have input schemas with typed parameters and descriptions. Tool descriptions explicitly state cost implications and prerequisites (e.g., 'call find_company first'). However, output schemas are not documented, LLMs cannot predict response structure for downstream chaining. Risk annotations (READ_ONLY, WRITE) are present but not formalized as toolAnnotations in the MCP schema. Parameter constraints (limit ranges, ticker format) are described in text but not enforced via JSON Schema patterns or enums.
Search the KRX directory of Korean listed companies. Free. Use this as the first step when you have a company name (English or Korean) but not the six-digit KRX ticker. Pass the resulting ticker to ``get_recent_filings`` (paid) or ``get_disclosure_summary`` (paid, when you also have a specific receipt number).
Fetch the AI-generated English summary of a Korean DART disclosure. **This tool spends real USDC from the configured wallet** — 0.005 USBC per call as of v0.1, settled on-chain via x402. The wallet pays only on a successful 200 response; 4xx/5xx failures do not settle.
Fetch the current per-endpoint pricing for koreafilings.com. This is a free call; it returns the x402 wallet address, network, USBC contract, and the price in USDC for each paid endpoint. Useful to confirm the payer will be settling on the expected chain before spending anything.
Fetch up to ``limit`` AI summaries for one Korean ticker. **This tool spends real USDC from the configured wallet** — 0.005 USBC × ``limit`` per call (default 0.025 USDC). The wallet pays only on a successful 200 response; 4xx/5xx failures do not settle. If you only have a company name, call ``find_company`` first to resolve the ticker.
Output schemas not documented. LLMs cannot predict response structure (fields, types, nested objects) for downstream tool chaining or data extraction. Responses from get_recent_filings, get_disclosure_summary, and list_recent_filings lack documented return types.
Parameter constraints (limit ranges 1-50, 1-100, 1-168; ticker format 6-digit) are documented in descriptions but not enforced via JSON Schema minValue/maxValue/pattern. LLMs may pass invalid values; validation happens server-side without guidance for self-correction.
Risk annotations (READ_ONLY, WRITE) are present in tool descriptions but not formalized as MCP toolAnnotations (readOnlyHint, destructiveHint). Clients cannot programmatically detect which tools modify state or cost money without parsing descriptions.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | C | 64 | 2026-07-28+ | v2 |
Browse recent DART filings across every listed Korean company. Free. Returns metadata only — no AI summaries — so an agent can decide which filings warrant a paid call. Each entry includes ``rcpt_no`` (for ``get_disclosure_summary``) and ``ticker`` (for ``get_recent_filings``).
No error recovery guidance. Descriptions warn about USDC costs but do not explain what happens on 4xx/5xx failures, how to retry, or how to verify payment settlement. LLMs lack actionable next steps on failure.
No pagination or result-limiting guidance for list_recent_filings. Description does not state max results returned, whether pagination is supported, or how to handle large result sets. Default limit=20 is reasonable but not documented as a cap.