The pricing layer, as an API.

Quickstart

Every market and provider page on Agentery.com is rendered from one of these calls — expand “See the MCP response behind this page” on any of them to see the exact payload.

curl -s https://agentery.com/api/mcp \
  -H 'content-type: application/json' \
  -d '{
    "jsonrpc": "2.0", "id": 1,
    "method": "tools/call",
    "params": { "name": "niche_report",
                "arguments": { "niche": "product-management-agent", "response_mode": "summary" } }
  }'

Examples

Common jobs, phrased the way a developer or founder would ask their assistant:

A. Choosing a capability (one call)

“I need an agent to reconcile supplier invoices — which products fit and what do they cost?” → research_capability finds the nearest matching products by meaning (no category lookup), with pricing context and a ready-to-compare shortlist in one call → compare_providers on the returned compare_ready handles. Prefer to drive retrieval yourself? search_providers returns the nearest products to any natural-language query and says so when nothing is a strong match.

B. Pricing a new service

“I’m launching a code-review agent. What do comparable Individual, Pro and Team plans charge?” → find_niche → price_benchmark (per provider type, buyer tier and unit).

C. Monitoring a niche

“Which customer-support agents repriced this week, and what moved the niche benchmark?” → niche_report (movement + named drivers).

D. Reading the index

“What is the current AEPI level and its 1-day, 7-day and 30-day change? Which buyer tier has inflated most this month?” → get_price_index. “Return the 90-day AEPI history for the Pro tier” or “How has customer-support-agent pricing moved this week?” → get_price_index_history. AEPI is a chained like-for-like price index, not a market price.

E. Your own private benchmark

“Benchmark just Intercom Fin, Zendesk AI, Ada and Decagon against each other” or “fork customer-support-agent, drop these two, add my competitor” → create_custom_benchmark returns a one-time secret cb_… token (no account needed — that token is your only key). Then “what’s the current benchmark for cb_…?” → get_custom_benchmark, and “show its 30-day history” → get_price_index_history with benchmark_id. Editing membership creates a new version and shows the exact effect on the median; pass as_of to reproduce a past date. All stats reuse the same immutable pricing engine as the public niches; nothing you do changes the canonical data.

Core concepts

Observed pricing

Prices parsed daily from each provider’s public pricing (schema.org JSON-LD + rendered pages), not self-reported. Each carries an observed_at timestamp.

Plan matching

Histories follow the same named plan over time (paren-stripped, punctuation-normalised), so a classifier change or a second plan sharing a tier can’t fake a move.

Provider type

Agents and MCP servers are two different markets that price differently, so they are benchmarked separately — a blended agent+MCP median is never the headline. Set provider_type when you know it; otherwise price_benchmark returns the per-type / per-tier matrix.

Buyer tiers

Individual, Pro, Team / SME and Enterprise are benchmarked separately — never blended into one number.

Index methodology

Chained and like-for-like: only agents priced on both consecutive scan days move the index (geometric mean of per-agent ratios), so revealing/hiding a price never fakes a market move.

Thin cohorts

Small samples are flagged loudly (confidence thin) rather than smoothed over.

Pricing recommendation

Positions a price against the observed p25 / median / p75 for a niche + tier, calculated from the same observed data; returns applicable:false for free / contact-sales / usage cohorts rather than inventing a monthly number. A dedicated MCP tool will be added separately.

Liveness

A daily endpoint probe; providers show uptime and consecutive-failure state.

Tool reference

19 tools. Each call is JSON-RPC tools/call. Inputs and outputs follow the endpoint’s tools/list schema. Writes are limited to report_outcome and the custom-benchmark tools — the latter only ever touch your own private cohort, never the shared pricing dataset.

Live examples: a niche_report page · a get_provider_profile page.

Built for current assistants — ready for autonomous agents

Today, humans usually initiate the research and approve the decision. Agentery lets Claude, Cursor and other agents perform the market analysis. The same structured pricing, confidence and provenance fields are designed to support increasingly autonomous procurement workflows as they emerge.

Methodology

Prices are observed daily from each provider’s public pricing — parsed, not self-reported. The index is chained and like-for-like; tier medians and quartiles are computed per buyer persona; thin cohorts are flagged; liveness is a daily probe. Nulls mean “not observed”, never zero: a missing price is returned as null with a reason, and aggregates exclude it rather than treating it as $0.

Access & limits

Authentication

None required for public reads.

Read / write

15 read-only tools. The 4 writes — report_outcome and the *_custom_benchmark tools — only add outcome signals or manage your own private cohort, never the shared pricing dataset.

Rate limits

Public calls are rate-limited per client; identify heavy or commercial use.

Errors

JSON-RPC errors carry a message and a hint (e.g. unknown niche → “call market_gaps to list valid slugs”).

Freshness

Responses include the freshest source-observation timestamp; the market is re-scanned daily.