Public case study · historical · no identifying data

How one customer-support search helped improve Agentery.

An external MCP client researched customer-support tools on 24–25 July 2026 (UTC). This page records the research sequence, the limitations it exposed and the resulting improvements. Prices shown belong to that historical snapshot (usage events of 24–25 Jul 2026 and price-history rows dated 24–26 Jul 2026), not to today's pages.

Try a similar task today Connect Agentery MCP →
What happened · 24–25 July 2026 (UTC)

One task became a market, a price context and a shortlist.

The client did not have to name a category first. Agentery resolved the task to a market, returned that market's observed price cohorts, and the later searches produced a shortlist that was compared in one call.

The session also exposed real weaknesses: the first search returned generic infrastructure listings, and one narrowed query (“customer support chatbot helpdesk ticket resolution”) returned nothing at all. Those failures changed the product afterwards.

The record shows research calls only. It does not show a purchase, a saving, or that any provider completed a task, and this page claims none of those.

01
Describe the buying problem

On 24 July the client asked broadly for customer-support agents (task “customer support”) and ran a first search for “customer support agent”. That first search returned generic infrastructure listings rather than support products.

02
Identify the market

Agentery resolved the free-text task to the customer-support (tier-1) market and returned that market’s observed price cohorts by buyer tier. The client also asked for a price benchmark; the requested slug did not exist and was resolved to the same market, which the response said explicitly.

03
Narrow the search

On 25 July the client ran four searches in a row: “customer support agent for small business”, “customer support chatbot helpdesk ticket resolution”, “customer support tier 1 automated responses” and “chatbot customer support website widget”. Three returned ten results each; the helpdesk-ticket query returned nothing.

04
Shortlist providers

The website-widget search surfaced the eventual finalists. Any budget or integration preference the client applied is not part of the recorded summary, so this page does not claim one.

05
Compare finalists

Five providers were compared in one call. The comparison returned each provider’s own observed price, billing basis, source page and observation date. It did not rank them against a market cohort — that is why no position label appears in the table below.

What the comparison returned at the time.

Each finalist's own observed plan on 25 July 2026: plan name, monthly figure, billing commitment and the source page the price was read from. Where no monthly plan was observed, Agentery said so rather than inventing a position.

Chat Thing
plan: Basic
$11.67/mo
billed yearly (annual commitment); free tier available
CrafterQ
plan: Solo
$29/mo
billed monthly; free tier available
Recursive Support
plan: Starting plan
$49/mo
billed monthly
Deskcrew Support
plan: PRO
$24.17/mo
$290/yr, shown as a monthly equivalent (annual commitment); free tier available
Wonderchat
plan: —
no monthly plan observed
insufficient comparables — no priced monthly plan to place
free tier; Enterprise “starts at $1,499”; message credits $29.99
Market context the client received on 25 July 2026 · customer-support (tier-1) market · buyer-tier cohorts, any billing unit
Individual tier
median $9/mo
39 priced of 47 listings
Pro tier
median $50/mo
35 priced of 36 listings
Team / SME tier
median $249/mo
31 priced of 32 listings

The comparison itself did not place the finalists in these cohorts, so no per-provider “below” or “above” verdict is shown. Two of the four monthly figures are annual-commitment prices expressed per month; a like-for-like position would need the same billing commitment on both sides.

What it changed

The failures shipped fixes.

  • Search became meaning-first: it resolves the buyer's market and ranks inside it, instead of matching keywords across the whole directory. A search that finds nothing now says so with a labelled envelope instead of a bare empty list.
  • Integration filters became soft preferences: an integration Agentery has not indexed relaxes the constraint and is labelled “unverified”, never counted as present. A requirement stays a requirement; an unknown integration is a possibility to check, not a match.
  • Comparisons state their pricing basis: a comparison now reports how many of the compared plans carry an observed numeric price, which are contact-sales, and which are unpriced, so a capability comparison is never mistaken for a price comparison.
Try a similar task today

Follow the current search, profile and comparison steps with your own requirements.

Available products and pricing will differ from the historical example above. Today's path is search_providers → get_provider_profile → compare_providers; the starter prompt asks for the budget, the actual billing commitment and confirmed integrations, with unknown or unsuitable candidates kept separate.

Starter prompt

“Find customer-support agents for a small business with a budget of $100/month. For each candidate give the observed price, its plan name and billing commitment (monthly or annual), and whether Zendesk, Intercom or Freshdesk integrations are confirmed. List candidates with unknown pricing or unconfirmed integrations separately, and do not treat an unknown integration as a match.”

Worked examples with recorded results are on the Connect page.

Claude Code
claude mcp add --transport http agentery https://agentery.com/api/mcp
Adds Agentery as a remote streamable-HTTP MCP server.
Published safely: no caller hash, gateway profile, precise timestamps, geography, account details or raw telemetry. The external client is an MCP caller; we do not claim it is a specific person or organisation. Historical figures were re-checked against the recorded evidence on 22 September 2026.