MCP — computed analytics, not search

Competitors ship semantic search over filings. We ship computed analytics: peer percentiles, state medians, group min/median/max LCM, expense component benchmarks, and deterministic briefings — derived from the same Postgres slice API that powers Site v2.

Streamable HTTP, plain JSON-RPC 2.0. POST only— one JSON response per request. Auth: Bearer token or ?key= (site cookie also works for browser clients on companion APIs).

POST https://serffsurf.com/api/mcp
Authorization: Bearer <your-access-code>
Content-Type: application/json

# or
POST https://serffsurf.com/api/mcp?key=<your-access-code>

Need a code? Request access. The same partner code that unlocks the site works here.

OAuth / DCR: discovery stub at /.well-known/oauth-protected-resource. Full OAuth is not implemented yet — keep using bearer / ?key.

1. Open Claude → Settings → Connectors (or Custom connectors).
2. Add a remote MCP server with URL: https://serffsurf.com/api/mcp
3. Set authentication to Bearer token and paste your SerffSurf access code (or append ?key=<code> to the URL).
4. Enable the connector in a chat and ask one of the example questions below.

1. Create or open a Custom GPT / use Apps & Connectors where remote MCP is available.
2. Point the MCP / action endpoint at https://serffsurf.com/api/mcp
3. Configure Authorization: Bearer <your access code>.
4. Confirm tools appear (market_summary, group_position, expense_components, …) then try an example question.

1. Open Cursor Settings → MCP (or edit ~/.cursor/mcp.json).
2. Add a server entry with transport type streamable HTTP / url pointing at the endpoint above.
3. Include header Authorization: "Bearer <your access code>" (or use ?key=).
4. Reload MCP servers and invoke a SerffSurf tool from Agent chat.

No data is shown on this page — ask your connected client so results come through the gated tools.

• “What’s the median WC LCM in Georgia, and how does taken compare to indicated?”

  Uses market_summary — computed median/quartiles + methodology + rate trend across filed carriers (~47 states), not document retrieval.

• “Where does the Travelers group sit vs peers — min, median, max LCM by state?”

  Uses group_position — member roster plus per-state LCM min/median/max and best (lowest-LCM) entity.

• “What’s the median commission provision by state?”

  Uses expense_components — one-call computed median/quartiles from filed expense provisions.

• “Where does Hartford sit on LCM in Illinois?”

  Uses carrier_snapshot / carrier_position — computed peer percentiles + recent rate takes.

• “Show regulator objections mentioning tier in Michigan.”

  Uses objections_search — structured objection letters with disposition and excerpts.

• “What are executives saying about workers’ comp right now?”

  Uses industry_sentiment / market_voices — named-executive SEC quotes by LOB.

• market_summary
  Computed median/quartile LCM + methodology + movers (not RAG; ~47 states)
• carrier_snapshot
  Computed multi-state peer percentiles + latest rate change
• carrier_position
  Computed percentile LCM rows + rate actions for one carrier
• group_position
  NEW — group members + per-state LCM min/median/max + best entity
• expense_components
  NEW — median expense provision by state (e.g. commission) in one call
• territory_compare
  NEW — ZIP territory relativities when territories table is loaded
• briefings
  Deterministic rule-engine observations (alias: briefing)
• objections_search
  Regulator objection letters + excerpts (alias: search_objections)
• industry_sentiment
  Executive earnings sentiment by LOB (alias: market_voices)
• search_facts
  Evidence-linked LCM/deviation facts with tracking ids
• mi_lab_search
  MI PPA / HO lab facts (experimental)
• data_overview
  Corpus counts + aggregate teaser stats