Read-only MCP server that lets an agent diff SEC filings section-by-section and answers only with verbatim, cited text.
Strong tool definitions with comprehensive schemas, clear descriptions, and excellent error handling guidance. All 7 tools have detailed descriptions (100-300 chars), well-defined input schemas with type constraints and patterns, and output schemas documented in code. Tool names follow verb_noun convention (resolve, list, get, diff, search). Pagination is properly implemented with offset/limit/maxChars controls. Citations are embedded in responses for traceability. Minor gaps: parameter descriptions could be more explicit about dependencies (e.g., when to use diff_all_items vs diff_sections), and no explicit enum constraints for form types.
Compare all Items between two filings and return statistics only (no paragraph text). Results are sorted by lowest similarity first, so the most-changed Items appear at the top. Use this to triage which Items deserve close reading.
Return paragraph-level changes between one Item in two filings. Each change (added, removed, changed, unchanged) carries full citations for both sides. Unchanged paragraphs are included by default; set onlyChanges to omit them. Raise maxChars or call again with offset = offset + returned for the next page.
Return a page of verbatim paragraphs from one Item, each with a citation. If the Item cannot be located the result is status "not_found" together with the Items that are available; never a guess. Short placeholder bodies ("None.") are returned with a warning. Raise maxChars or call again with offset = offset + returned for the next page.
List recent filings for a CIK, optionally filtered by form ("10-K", "10-Q"). Each entry includes the accession number needed by other tools.
Parse a filing and list the Items that were found (e.g. 1A Risk Factors, 7 MD&A) with sizes and any parser warnings. Call this before get_section when unsure what exists.
Form type parameter in list_filings lacks enum constraint; description says 'Exact form type, e.g. "10-K"' but does not enumerate valid values (10-K, 10-Q, 8-K, etc.). LLMs may hallucinate invalid form types.
Item parameter across get_section, diff_sections, search_filing accepts flexible formats ('1A', 'Item 1A', 'II.1A') but does not document the full grammar or validation rules. LLMs may pass invalid item references.
No explicit guidance on when to use diff_all_items (statistics only) vs diff_sections (paragraph-level changes). Descriptions are clear but lack a 'when to use' hint to guide LLM tool selection.
Pagination logic is correct but maxChars truncation reason messages could be more actionable. 'entry at offset X alone exceeds maxChars' tells the LLM to raise maxChars but not by how much or what the entry size is.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | C | 64 | 2025-06-18+ | v2 |
Look up a company by ticker (exact) or name (substring) and return CIK candidates. Returns an empty results array when nothing matches. A ticker can point at a newly created holding company with no 10-K/10-Q yet; if list_filings returns nothing for a form, search again by company name and check the other CIKs.
Search for a regex pattern within one Item of a filing and return all matching paragraphs verbatim, each with a citation. Matches are returned in document order. Raise maxChars or call again with offset = offset + returned for the next page.
INSTRUCTIONS constant documents the tool flow but is not exposed to the LLM as a system prompt or resource. Agents must infer the workflow from tool descriptions alone.