For AI agents

Everything an agent needs to decide, with its operator, whether a skill here is worth it.

This is a commercial registry. It sells some of the skills listed here, through Polar as merchant of record, and its operator earns money when one is bought. The evaluate answer can be "do not buy": it says so when the measured evidence does not cover your task, when the price is not recovered, or when a free alternative is enough. Every purchase ends in a checkout that a human completes.

Connect

MCP (Streamable HTTP): https://agentickeychain.com/mcp
MCP protocol versions: 2026-07-28, 2025-11-25, 2025-06-18, 2025-03-26, 2024-11-05
HTTP API: https://agentickeychain.com/v1/ (service description)
OpenAPI: /openapi.json
Skill index: /skills/index.json, /skills/index.md
Signing keys: /keys.json
Agent manifest: /.well-known/agentic-keychain.json
Claude Code plugin: /plugin marketplace add https://agentickeychain.com/integrations/claude-code/marketplace.json, then /plugin install agentic-keychain@agentic-keychain. Free. It adds this MCP endpoint and the access skill.
Agent Skill: SKILL.md, the free access skill, listed in /.well-known/agent-skills/index.json

Tools

resolve_task — Describe a task in plain words and get one compact answer: whether a registry capability is worth using for it, from registry-run benchmarks only, and the next step.
evaluate_capability — Decide whether a registry capability is worth using or buying for your task, from registry-run benchmarks only.
search_capabilities — Search capability names, summaries, tasks and use-when conditions by keywords, ranked by lexical relevance only, never by payment.
compare_capabilities — Assess two to five named capabilities side by side for the same inputs, with the same method and ak.offer.v1 offer shape as evaluate_capability.
estimate_roi — Per measured model: expected saving per run, break-even runs and net saving over expected_runs, with intervals, for one capability and no task.
get_capability — The capability's card: purpose, use-when and do-not-use-when conditions, permissions, price, evidence summary, trust checks, links and a task-independent ak.offer.v1 offer.
get_benchmark — Registry-run benchmark summaries (labelled), publisher claims (labelled PUBLISHER CLAIMED) and the signed attestations of one capability.
get_provenance — Artifact hash, manifest hash, signed publisher and registry attestations and a transparency-log inclusion proof for one capability.
get_quote — A signed 15-minute price quote with the hosted checkout link.
unlock_capability — Exchange the license key from the checkout (a purchase requires operator approval) for a signed entitlement and a download link that lasts at most 15 minutes.
report_outcome — Report how a capability worked for you.

Start with resolve

POST a plain description of the task to resolve_task and get one compact answer: a decision word, the reason codes, the measured saving per run with its interval (or null), the confidence, the evidence label, the price and the next step. Any field it does not read is refused with the list of fields it does read.

POST https://agentickeychain.com/v1/resolve
Content-Type: application/json

{
  "task": "Summarize a 40,000-line application log and group its recurring errors",
  "model": "claude-sonnet-5"
}

The decision is one of use, do_not_use, use_free_alternative, insufficient_evidence, blocked_by_constraints, no_match. do_not_use and no_match are normal answers. The same inputs sent to evaluate_capability give the full arithmetic, the offer and the operator_message.

Full evaluation

POST a plain description of the task you are about to do. Add your model and, if you know them, how often you do this task and what one run costs you now; without a baseline the saving for your task stays unknown and the answer will not recommend a purchase on it.

POST https://agentickeychain.com/v1/evaluate
Content-Type: application/json

{
  "task": "Summarize a 40,000-line application log and group its recurring errors",
  "model": "claude-sonnet-5",
  "expected_runs": 20,
  "baseline_cost_estimate_usd": 0.6
}

The decision is one of recommend, do_not_recommend, undetermined or no_relevant_capability. Each answer lists its reasons, the evidence label behind every number, free alternatives and an operator_message written for your operator.

Buying

Ask for a quote: POST /v1/quote with the capability id. The quote is information for your operator, valid 15 minutes; it is not a payment.
Your operator completes the hosted checkout. You never enter payment details.
Polar issues the license key with the purchase; it is listed on the purchases page of Polar's customer portal. Post it to /v1/unlock. You get an entitlement signed with the registry's entitlement key and a download link that lasts at most 15 minutes.
Verify the zip's SHA-256 against the entitlement before you install it.

Access follows the license status at Polar. A license check is reused for at most 24 hours; entitlements issued on it expire with it, and a download link lasts at most 15 minutes. Once Polar no longer reports the license as granted, unlock is refused from the next check on. POST /v1/unlock again with the same key at any time for a new link; it never charges.

Paid capabilities are sold by Polar Software, Inc., 3500 South DuPont Highway, Dover, DE 19901, USA, as merchant of record and reseller. They are published by the registry's operator, Vayu Solutions, van Pijkerenlaan 28, 8096 VE Oldebroek, the Netherlands, under its first-party publisher name Agentic Keychain, and licensed by Vayu Solutions to the buyer. Consumers in the European Economic Area can withdraw within 14 days, also after downloading: /policies/refunds/v1/.

The prices of some capabilities can change, at most once a week per capability. A change applies to every buyer at the same time; no buyer gets a price of their own. You pay the price shown in the checkout you complete. A quote shows the price in force when it was issued and is not binding. A later change does not alter what an earlier purchase cost and does not lead to a refund of any difference; the refund policy applies as it stands.

Refund policy (version 1, SHA-256 895f3ece923e48e4…): /policies/refunds/v1/

Reporting how it went

After using a skill you may call report_outcome. Reports are stored as AGENT REPORTED, never change evidence labels, trust or ranking, and no task text is kept.

Ranking

Evaluate ranks by: decision; lower-bound net saving; relevance; capability id. Search ranks by: lexical relevance (BM25 over name, summary, purpose, tags, tasks); capability id. Never used for ranking: payment, bid, seller subscription, first party, agent reported outcomes, checkout availability, price hold.