OpenFilings provides 7 read-only tools with consistent naming (verb_noun pattern: companies_search, filings_list, etc.) and reasonable descriptions (avg ~120 chars). All tools have input schemas with types and descriptions. However, output schemas are not documented in the visible code, responses are shaped via helper functions (company_summary, filing_summary) but the LLM cannot see the structure. Parameter descriptions are present but often terse (e.g., 'Data source filter' lacks guidance on valid values). No error recovery guidance, no tool annotations (readOnlyHint missing despite all being read-only), and no pagination metadata (total count, next_cursor) visible in responses. Baseline: avg tool description ~120 chars (within 10-1024 range), all params typed, but output structure opaque to the LLM.

• Document output schemas for all tools. Add a 'Returns' section to each tool docstring specifying the structure of success() responses (e.g., 'Returns {companies: [{id, name, ticker, source}], count: int, next_steps: str}').
• Enumerate valid values for 'source' parameter across all tools. Replace 'Data source filter' with 'Data source (all, fca_nsm, cvm, edinet, sec, etc.). Defaults to all.'
• Add readOnlyHint=true to all 7 tool definitions via @mcp.tool(readOnlyHint=True) to signal safety to LLMs.
• Complete the historical_backfill description. Explain what 'historical' means (how many periods?), when to use it vs company_facts, and what structure it returns.
• Enhance error messages with recovery guidance. E.g., 'Company not found. Try companies_search() with a partial name.' instead of bare exception text.