The Swallow SERFF MCP server is a Model Context Protocol endpoint that turns the indexed SERFF filings corpus into a conversational research surface for LLM clients including Anthropic Claude Desktop, Claude.ai, OpenAI ChatGPT, Cursor, and Claude Code.
In one line: paste one URL into your LLM, sign in, and ask actuarial questions in plain English. The model searches filings, reads summaries, walks predecessor chains, and pulls structured rate tables on your behalf with no SQL, downloading PDFs, SERFF portal trawling, or additonal code dependency.
Model Context Protocol is an open standard originally published by Anthropic Anthropic that lets LLM clients call external tools through a single JSON RPC interface. The major LLM vendors have converged on MCP, including Claude, ChatGPT, Cursor, and Gemini. An MCP is like a plugin or connector to you existing LLM client experience.
Anyone whose job already involves reading, analyizing and submitting SERFF filings.
Typical users include:
• US admitted P&C carriers running competitive and pricing intelligence
• MGAs and programme managers diligencing incumbent rating plans
• Reinsurers tracking direct and fronted programme behaviour
• Captives and risk retention groups benchmarking against the marketInsur
• Tech new entrants researching state by state competitive landscapes
• Actuarial consultancies pulling filing precedent for client deliverables
• Compliance and state filings teams researching DOI objection patterns
Typical roles inside carriers include:
• Pricing actuaries
• Product managers
• State filings analysts
• Underwriting leadership
• Competitive intelligence teams
The SERFF portal is effectively a search box attached to a document repository. Finding a filing usually means downloading large PDFs, manually reading them, repeating the process across carriers, and trying to track changes by memory. The MCP server collapses that process into a single conversational workflow.
Examples include:
• “List every California auto filing in 2025 that added a telematics surcharge”
• “Walk this Progressive programme back to its ISO bureau root”
• “Pull the rate manual XLSM out of REGU-134742228”
• Competitive rate scans that previously took a day become a single prompt
• Programme lineage walks reduce hours of manual cross referencing
• Filing triage becomes near instant
• Source documents can be pulled directly without opening the portal
• Every LLM claim remains traceable to SERFF ids and page references
• Conversations themselves become reusable and sharable workpapers
The platform indexes the full SERFF filings record, including:
• Rate filingsRule filings
• Programme filings
• Form filings
• Correspondence documents
Coverage today includes California from 2005 onward, alongside bureau and predecessor chains including ISO, NCCI, AAIS, and MSO. Additional states are onboarding through 2026.
If it has been filed in SERFF, it is included.
The MCP server acts as the reading and reasoning layer over SERFF filings.
The wider Swallow platform extends this further by turning filings into:
• Live rating APIs
• Product monitoring
• Market pricing benchmarks
The MCP can be used independently as a research tool, or combined with Swallow Rating, Filing Assistance, Monitoring, and Market Pricing Analytics for execution workflows.
Learn more at Swallow.app
The Free tier includes:
• Full structured filing search
• Semantic summary search
• Filing summaries
• One user seat
Learn how to set it up here
Pro adds the deep research surface, including:
• Paragraph level semantic search
• Structured extract JSONs
• Full history reconciliation
• Source document downloads
• Multi user seats
If Free answers “what changed?”, Pro answers: “Show me the rate tables, factor curves, rating equations, and XLSM rater.”
Every filing is embedded into vector representations.
• Summary search uses approximately 65K filing level embeddings
• Deep search uses 12.4M paragraph level embeddings
Natural language queries are embedded and cosine compared directly against the index.
Benefits include:
• Millisecond search times
• Low compute cost
• Discovery beyond keyword matching
Each filing can expose eleven extracted artefacts:
• rates_data - rate tables
• calculations - rating walk throughs
• final_rating_calculation - canonical rating expressions
• examples - worked premium examples
• coverages - perils and limits
• deductibles - deductible rules
• discounts - discounts and surcharges
• endorsements - optional riders
• exclusions - exclusions and conditions
• forms - policy forms and editions
• underwriting_guidelines - underwriting rules
Yes. For Free you get summaries of filing content which augmenets LLM conversational logic.
For PRO, the corpus supports:
• Class codes
• Experience mods
• Territory multipliers
• Deductible factors
• Protection classes
• Endorsement pricing
• Minimum premiums
• State taxes
• Carrier deviations from ISO, AAIS, and NCCI
Yes. Key improvements include:
• Conversational workflows instead of portal navigation
• Cross filing reasoning
• Reconciled filing history
• Fast semantic search
• Structured filing extraction
• One endpoint across all major LLM clients
Open Claude Desktop, Claude Web, or Claude Code and:
1: Go to Settings → Connectors → Add custom connector
2: Paste: https://api.serff.ai/mcp3: Complete authentication
The tools will then appear directly in the LLM client.
Setup guide:
How to connect the Swallow SERFF MCP server to Claude
In ChatGPT:
1: Open Settings and activate developer mode
2: Open Connectors
3: Add a custom connector
4: Paste the MCP URL
5: Complete authentication
The tools become available inside ChatGPT conversations and Deep Research workflows.Setup guide:
How to connect the Swallow SERFF MCP server to ChatGPT
Authentication uses OAuth 2.1.
The client:
1: Connects to the MCP endpoint
2: Receives the authentication metadata
3: Opens the Swallow sign in flow. You will receive auth details on signup and recoverable with Swallow -> Settings.
4: Receives a scoped bearer token
Yes. Create a free account at: Swallow account signup
The same account works across both Free and Pro. You just need to upgrade to a PRO account.
Ask the model to call:
• mcp_health
The response includes:
• Build SHA
• Server version
• Tool surface
• Corpus watermark
• Tools hash
Most LLM clients support MCP server plugins or connections, it can usually integrate successfully.
The MCP server is plain JSON RPC 2.0 over HTTPS. It is stateless rather than SSE.
Yes. You can download source files via signed URLs which have an expiry of 15 mins. You can always request again if the link expires.
Yes. The same surfaces powering internal tooling are available externally:
• Structured search
• Semantic search
• Structured extracts
• Source document downloads
Paste one MCP server URL (https://api.serff.ai/mcp) into a LLM client of your choice - Claude, ChatGPT, Cursor, Gemini etc. Sign in, and the indexed SERFF filings corpus becomes conversational.
Search, summaries, lineage, structured rate tables, and source documents all become accessible through natural language. If you read SERFF for a living, Swallow is built for you.
Yes. No mutations are possible through the MCP surface.
Yes. The ingestion pipeline continuously processes new SERFF filings and republishes the indexed corpus within 24 hours.
No. The server is deterministic. It only returns filed values. The MCP server will not invent anything. If the corpus lacks a value, the model points users back to the source documents instead.
Current indexed coverage includes California from 2005 onward. Additional states are onboarding throughout 2026.
Yes, on PRO. No, on Free.
No. Just an annual subscription if you are on PRO or Custom.
Yes, for a week, but you need to contact us first.
