MCP App Store
by KeenEthics
MGT Quoting icon

MGT Quoting MCP Server for Claude and ChatGPT

By MGT Insurance

Overview

MGT Quoting lets licensed insurance agents quote MGT small-business commercial insurance without leaving Claude. Describe the business in plain language, and Claude classifies it, checks it against MGT's appetite, and builds the quote with you. It covers all products you can quote through the MGT platform. Claude works through MGT's quoting steps for the product you ask for. It classifies the business, asks only for the details that product needs, creates a draft quote, and validates it before rating. You can change limits, coverages, or business details in conversation, and Claude re-validates and re-rates the quote. Quotes are saved to your MGT agency account, so you can pick them up in the MGT agent portal. Claude doesn't bind coverage or take payment. Every quote stays subject to MGT underwriting review. You need an MGT agency account with quoting access; sign in with your existing MGT credentials. Quoting output is not insurance advice. For licensed insurance professionals only.

How to connect

  1. Claude: Open the MGT Quoting listing in the Claude connectors directory and choose Connect. Claude listing MGT Quoting exposes 8 tools here.
  2. ChatGPT: Open the MGT Quoting listing in the ChatGPT app directory and choose Connect. ChatGPT listing MGT Quoting exposes 8 tools here.

Tools

appetite_check

ChatGPT
Run an appetite check for a business. Returns per-product decisions (available | moratorium | declined). Resolves eligibility one of two ways: pass an exact naics code (evaluated hierarchically — 4, 5, or 6 digits all resolve directly, no truncation), or a free-text businessDescription to have it classified automatically against the NAICS 2022 catalog — see the businessDescription field description for the resulting candidates response field. Prefer the exact naics field when known. A naics that is real but not specific enough to resolve on its own (e.g. 5311 "Lessors of Real Estate") is REJECTED with a 4xx — see appetite/appetite_naics_expand_rejection — instead of returning product decisions; retry with one of the rejection's named candidate codes. The moratorium arm of the check is location-precise: pass line1+city (point check) or at least postalCode (conservative ZIP boundary-overlap) whenever the risk address is known — with only a state the moratorium check is skipped and the decisions carry no moratorium signal.

create_quote

ChatGPT
Create a specialty insurance quote for one or more products. Accepts partial information — returns a rate when sufficient data exists, or lists required fields otherwise. POSTs to /api/internal/quotes. Supported products: bop, specialty-gl, specialty-property, farm, om-bop, om-auto, om-farm, om-liability. Quote attribution (organization, owner, Ascend commission routing) is resolved server-side from the calling operator's verified identity — this tool does not accept and never advertises identity fields. The one selection input is the optional organization, used only by a caller who belongs to more than one organization: when it is needed and omitted, the tool returns status needs_organization with organizations[] ({id, name}) and creates nothing — ask the user which organization to use, then call again with organization set to the chosen id; the server refuses an organization the caller is not a member of. Enum/options field values are validated against the product schema; an out-of-options value is REJECTED (HTTP 400) and no quote is created — call get_product_create_schema first to discover the valid options for each field before setting it.

edit_quote

ChatGPT
Edit an existing draft quote in Shadowfax: set field values on ANY element and, optionally, re-rate. POSTs /api/internal/quotes/apply-field-values, which runs the shared @mgt/quote/edit kernel (guard -> reset-if-finalized-and-rating -> validate each field against the product question set -> auto-create missing path elements -> grouped write -> optional re-rate). This ONE tool replaces the old modify_quote (root scalar edits), add_location, and add_building tools: a single call can set root fields, set fields on nested exposures, AND create the Location/Building elements it addresses. Each update is addressed by EITHER an explicit element_locator OR a type-indexed element_path (e.g. "" for the product root, "Location[0]", "Location[0].Building[1]"); a path that doesn't exist yet is auto-created parents-first (element types are validated against the product question set, so a bad type is refused, not created). Each update value is always a string: send scalar values as their textual representation, and JSON-encode the declared object or array shape into a string for structured group or multivalue fields. The edit kernel parses those JSON-encoded strings before writing; do not send a native object or array directly. Fields that aren't a form question on the target element are skipped (not written) and reported in skipped -- never a hard failure. An enum/options field value that is not one of the schema's allowed options is likewise validated and rejected -- reported in skipped, left in required_to_rate (not applied); call get_product_create_schema first to obtain the valid options for a field before setting it. Set rate: true to re-rate after writing; after a clean preflight, pass an explicit empty updates array with rate: true to rate without rewriting an arbitrary field. This is the only empty-update form: omitting rate or setting it false requires at least one update. The result then carries rating, which is ONE OF THREE shapes: on a rated response { status: 'bindable' | 'referred', premium, bindable, rate_refinements } -- bindable means the quote can be bound as-is, referred means it rated and produced a real premium but underwriting sent it to a human for review, so it cannot be bound or edited further (rate_refinements are optional fields that would sharpen the rate, and are always [] on referred); on an incomplete response { status: 'needs_input', required_to_rate } (fields that must be provided before a rate can be produced); or, when underwriting refuses the risk outright, { status: 'declined', decline_reasons } — a list of why. A decline can be CAUSED by the edit you just made (answering that a contractor uses subcontractors, for example) and is TERMINAL: there is no premium, no field you can supply to reverse it, and every further edit to that quote is refused as not-draft. Relay the decline and its reasons to the human rather than retrying with different values. A rated response may also carry side_effects.effective_date { from, to } (YYYY-MM-DD, the calendar days the quote page shows): the re-rate rolled the quote's stale effective date forward, which also moves the term start and can apply the default coverage terms for the new date. Tell the human the date moved and to what; it is absent when the date did not move. Without rate, the write-only result is { success, product, appliedCount, skipped, created, required_to_rate, rate_refinements }. required_to_rate and rate_refinements are present on EVERY response regardless of rate — each a list of field descriptors ({ name, friendly_name?, label?, type?, owningMember, owningElement }); required_to_rate lists fields still needed to rate, rate_refinements lists fields that would sharpen the rate. Both default to [] (the transaction/MTA segment path never computes them). Only edits DRAFT quotes on the write-only path; a finalized quote is reset first when rating, and refused (not-draft) when write-only.

get_product_create_schema

ChatGPT
Return the CREATE-payload schema for a product: a nested template for ONE products[i] entry's fields — arrays for /+ quantified exposures/composite fields (e.g. BopLocation, each holding a BopLocationBuilding array), plain nested objects for singular exposures/composite fields (e.g. Liability, industry_group), and raw Socotra config field names (bop_/Bop_*, no friendly-name translation). Coverage terms/fields flatten onto their nearest containing exposure/product object rather than nesting under the coverage's own name, matching the real create wire. Every leaf field/term descriptor also carries a path: its exact location in the create_quote payload for this product, rooted at products[0] (e.g. products[0].BopLocation[0].Property_deductible) — write it there verbatim rather than inferring nesting depth from the returned shape. A leaf also carries question — the broker-facing question text — only when it differs from label (omitted when absent or equal); use it to match a stated document value to a field. A composite field (list or nested object) cannot carry its own wording in body, so its label and question are returned in the response's top-level compositeFields map, keyed by the composite field's path. The returned body is the field template ONLY — it is not itself a valid create_quote call. To call create_quote, wrap it as one products[] entry with the 'product' discriminator: { products: [ { product: 'bop'|'specialty-gl'|'specialty-property', ...body } ], business_name, address, ... } — the discriminator and the account-level envelope (business_name/address/etc.) are supplied by the caller, not part of body. This is the CREATE-grammar sibling of get_product_quote_schema, which instead speaks generate_quote's flat patch-path grammar. Call this before create_quote to discover a product's field names, types, required-ness, and enum options, without reading config entities one by one.

get_quote_schema_catalog

ChatGPT
Parameters for the get_quote_schema_catalog tool. Returns the common (product-agnostic) field registry plus the list of products live for the calling user and their organization. This is a MAPPING REFERENCE ONLY: use it to map your own data onto quote fields, not to drive or replicate the create API's own required-vs-optional validation, which remains the quote API's job at submission time. Call get_product_quote_schema per product for the full per-product field/term list.

get_quoting_plan

ChatGPT
Parameters for the get_quoting_plan tool. Takes no input. Returns the step-by-step plan for quoting a risk with the quoting tools: which tool to call in which order, how to read each result, and how to hand the finished quote to the user. Call it first, before any other quoting tool, and follow it.

resolve_business_classification

ChatGPT
Resolve taxonomy relationships for one observed business-classification fact. Provide exactly one of naics, businessDescription, mgtBopClassCode, or industryGroup. This tool maps classification relationships only: it does not determine appetite, quote approval, product availability, or a decline.

validate_quote

ChatGPT
Validate an operator-owned draft quote without rating or underwriting it.

appetite_check

Claude

create_quote

Claude

edit_quote

Claude

get_product_create_schema

Claude

get_quote_schema_catalog

Claude

get_quoting_plan

Claude

resolve_business_classification

Claude

validate_quote

Claude

Similar Apps

Listed on MCP App Store

Listed on MCP App Store

Add this badge to your site or README. It links to this page.

HTML

<a href="https://mcp-app-store.keenethics-labs.com/app/mgt-quoting"><img src="https://mcp-app-store.keenethics-labs.com/badges/listed-on-mcp-app-store.svg" alt="Listed on MCP App Store" width="220" height="44"></a>

Markdown

[![Listed on MCP App Store](https://mcp-app-store.keenethics-labs.com/badges/listed-on-mcp-app-store.svg)](https://mcp-app-store.keenethics-labs.com/app/mgt-quoting)