Agent registries

# Not live yet (gaps #2)
curl -X POST "https://api.sb/leaseline.app/campaigns?mode=try" \
  -H "Authorization: Bearer $STARTUP_BUILDER_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{"play":"agent-registries"}'

 1import { $ } from 'startup-builder'
 2
 3return await $.Campaign.create({ play: 'agent-registries' }, { mode: 'try' })   // dry run on the day-0 Startup, 2026-08-24 · Not live yet, gaps #2

npx startup-builder do Campaign.create --json '{"play":"agent-registries"}' --try   # Not live yet (gaps #2)

{
  "name": "do",
  "arguments": {
    "code": "return await $.Campaign.create({ play: 'agent-registries' }, { mode: 'try' })   // dry run on the day-0 Startup, 2026-08-24 · Not live yet, gaps #2",
    "mode": "try"
  }
}

 1{
 2  "type": "https://startupbuilder.dev/errors/refused",
 3  "title": "Refused",
 4  "status": 422,
 5  "detail": "agent-registries can’t start yet: 1 check fails and 1 warns.",
 6  "checks": [
 7    {
 8      "check": "use-when",
 9      "status": "pass",
10      "observed": "the MCP server answers on leaseline.app; 2 of 3 Tools try with a sandbox key",
11      "doc": "/plays/agent-registries#check-use-when",
12      "then": "Metric.query"
13    },
14    {
15  …

curl "https://api.sb/leaseline.app/events?where=name%20in%20(%22Startup.advanced%22%2C%20%22Version.deployed%22%2C%20%22Monitor.checked%22%2C%20%22Action.evaluated%22%2C%20%22Message.received%22%2C%20%22Interface.called%22)&limit=1" \
  -H "Authorization: Bearer $STARTUP_BUILDER_API_KEY"

# Not live yet (gaps #7)
curl -X POST "https://api.sb/leaseline.app/triggers/fire?mode=try" \
  -H "Authorization: Bearer $STARTUP_BUILDER_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{"like":"event_1409"}'

# Not live yet (gaps #7)
curl "https://api.sb/leaseline.app/decisions?where=on%20%3D%20%22tool_3%22" \
  -H "Authorization: Bearer $STARTUP_BUILDER_API_KEY"

# Not live yet (gaps #7)
curl "https://api.sb/leaseline.app/actions?where=on%20%3D%20%22tool_3%22" \
  -H "Authorization: Bearer $STARTUP_BUILDER_API_KEY"

 1import { $ } from 'startup-builder'
 2
 3const { events: [last] } = await $.Event.list({ where: 'name in ("Startup.advanced", "Version.deployed", "Monitor.checked", "Action.evaluated", "Message.received", "Interface.called")', limit: 1 })   // { play } compiled to this Play’s Trigger Events
 4const { subject } = await $.Trigger.fire({ like: last.$id }, { mode: 'try' })   // the same Event again, in try mode · Not live yet, gaps #7
 5const { decisions: [decision] } = await $.Decision.list({ where: `on = "${subject}"` })   // what it chose, and why · Not live yet, gaps #7
 6const { actions } = await $.Action.list({ where: `on = "${subject}"` })   // what it did, with its checks · Not live yet, gaps #7
 7return { decision, actions }

npx startup-builder do Event.list --json '{"where":"name in (\"Startup.advanced\", \"Version.deployed\", \"Monitor.checked\", \"Action.evaluated\", \"Message.received\", \"Interface.called\")","limit":1}'
npx startup-builder do Trigger.fire --json '{"like":"event_1409"}' --try   # Not live yet (gaps #7)
npx startup-builder do Decision.list --json '{"where":"on = \"tool_3\""}'   # Not live yet (gaps #7)
npx startup-builder do Action.list --json '{"where":"on = \"tool_3\""}'   # Not live yet (gaps #7)

{
  "name": "do",
  "arguments": {
    "code": "const { events: [last] } = await $.Event.list({ where: 'name in (\"Startup.advanced\", \"Version.deployed\", \"Monitor.checked\", \"Action.evaluated\", \"Message.received\", \"Interface.called\")', limit: 1 })   // { play } compiled to this Play’s Trigger Events\nconst { subject } = await $.Trigger.fire({ like: last.$id }, { mode: 'try' })   // the same Event again, in try mode · Not live yet, gaps #7\nconst { decisions: [decision] } = await $.Decision.list({ where: `on = \"${subject}\"` })   // what it chose, and why · Not live yet, gaps #7\nconst { actions } = await $.Action.list({ where: `on = \"${subject}\"` })   // what it did, with its checks · Not live yet, gaps #7\nreturn { decision, actions }",
    "mode": "try"
  }
}

 1{
 2  "decision": {
 3    "$id": "https://api.sb/leaseline.app/decisions/decision_476",
 4    "$type": "Decision",
 5    "choice": "refresh-it",
 6    "reason": "version 1.4 renamed draft_notice to draft_renewal_notice, and 4 listings still show the old name",
 7    "stakes": "low",
 8    "options": [
 9      "list-it",
10      "refresh-it",
11      "test-a-description",
12      "fix-the-first-call",
13      "answer-the-reviewer",
14      "park-it"
15  …

action_3 resubmits the 4 stale listings, one registry at a time, after a 2-minute hold you can stop; each install string carries its registry tag.

≈ +$190 ARR a month at 300 views

Worth

30 min week 1, then 5 a week

Your time

$0.10 per listing refresh, ~$3 a month

Spend

Day 4: first call from a listing

First signal

an Interface whose first call works, 1+ Tool a stranger can try

Use it when

No first call works: agent-legibility · no Tool tries without a key: grant one (Offer.grant)

Not yet when

Worth’s formula: about +2.3 first calls per 100 listing views (Other Startups’ results; no lift of your own yet) × 300 views a month × 8 % first calls to paid × $347 first order ARR ≈ +$192. Under First signal: At 300 views a month, 2 Readouts of 28 days tell a registry at 2 per 100 from one at 0; below 100 views a registry reads views unreported beside Other Startups’ results.

Start it this weekCopy Anchor Link

npx startup-builder add play agent-registries   # 1 · writes plays/agent-registries.mdx (Not live yet, gaps #5)

# 2 · dry run; below, its 422 on the day-0 Startup (clock at 2026-08-24)
curl -X POST "https://api.sb/leaseline.app/campaigns?mode=try" \
  -H "Authorization: Bearer $STARTUP_BUILDER_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{"play":"agent-registries"}'

# 3 · keep it, at You approve each one for the first 9 listings
curl -X POST https://api.sb/leaseline.app/campaigns \
  -H "Authorization: Bearer $STARTUP_BUILDER_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{"play":"agent-registries"}'

# on
curl -X POST https://api.sb/leaseline.app/processes/launches/enable \
  -H "Authorization: Bearer $STARTUP_BUILDER_API_KEY"

# off: no submission or refresh goes; live listings stay
curl -X POST https://api.sb/leaseline.app/processes/launches/pause \
  -H "Authorization: Bearer $STARTUP_BUILDER_API_KEY"

{
  "type": "https://startupbuilder.dev/errors/refused",
  "title": "Refused",
  "status": 422,
  "detail": "agent-registries can’t start yet: 1 check fails and 1 warns.",
  "checks": [
    {
      "check": "use-when",
      "status": "pass",
      "observed": "the MCP server answers on leaseline.app; 2 of 3 Tools try with a sandbox key",
      "doc": "/plays/agent-registries#check-use-when",
      "then": "Metric.query"
    },
    {
      "check": "baseline",
      "status": "warn",
      "observed": "0 listings; 9 first calls from unknown callers in 90 days",
      "doc": "/plays/agent-registries#check-baseline",
      "fix": "n 9: the leaf has no Before; the parent reads before-after.",
      "then": "Metric.query"
    },
    {
      "check": "approved-pages",
      "status": "fail",
      "doc": "/plays/agent-registries#check-approved-pages",
      "fix": "Publish the Tools manifest Page; descriptions state only what it says.",
      "then": "Site.publish"
    },
    {
      "check": "role-unstaffed",
      "status": "pass",
      "observed": "Marketing Lead and Product Lead are your agents; reviewer asks come to you",
      "doc": "/plays/agent-registries#check-role-unstaffed",
      "then": "Role.staff"
    }
  ]
}

Listing is free on every registry Leaseline uses; each drafted description or refresh costs $0.10 from your Domain (Worker cost), and the install test runs in a sandbox at $0.02. That comes to about $3 a month at 9 listings refreshed twice.

When it startsCopy Anchor Link

{
  "$id": "https://api.sb/leaseline.app/events/event_1409",
  "$type": "Event",
  "name": "Version.deployed",
  "at": "2026-10-19T15:41:02Z",
  "seq": 1409,
  "data": {
    "version": "v1.4.0",
    "tools": {
      "renamed": [
        {
          "from": "draft_notice",
          "to": "draft_renewal_notice"
        }
      ]
    },
    "listings": {
      "stale": 4,
      "of": 9
    }
  }
}

# each delivery is signed with the Watch’s secret
curl -X POST https://api.sb/leaseline.app/watches \
  -H "Authorization: Bearer $STARTUP_BUILDER_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{"selector":"Version.deployed","url":"https://leaseline.app/hooks/sb"}'
npx startup-builder trigger Version.deployed --like event_1409   # replays the rename
npx startup-builder clock advance --by 7d   # runs the weekly crawl, which reads every listing back

Who runs itCopy Anchor Link

Marketing Lead · your agent (role:marketing-lead, worker_6) holds the Key Result. It hands failing first calls to Product Lead · your agent (role:product-lead, worker_8) and every reviewer ask to you (Dana Okafor, person, worker_1). Your number links to Did it work.

What it reads · What it may call. It reads your Tools manifest, each registry’s listing and its submission form, as data; it submits only through each registry’s own form or API, and a reviewer’s note never chooses what it writes. Its tools are POST /{namespace}/integrations/{integration}/call, PATCH /{namespace}/tools/{id}, POST /{namespace}/experiments/{experiment}/start, POST /{namespace}/issues, POST /{namespace}/tasks/{task}/assign, and any other call becomes an Approval under the Role’s Mandate.

You hold the Marketing Lead Key Result: 3 first calls per 100 listing views by 2026-12-31. ← worker, metric
On each Version, crawl or registry notice, choose one option and say why.        ← triggers, chooses
A description says only what the manifest says the Tool does.                    ← copyRules 1, lint-block
Say that send_renewal_notice needs a claimed key.                                ← your file, line 4; side-effect-hold
At most 3 new registries a week.                                                 ← your file, line 5

The slow loop, as a Change. decision_452 (2026-10-05, weekly) read 2 descriptions on the most-viewed registry: the one that opens with the task (“Lists leases due for renewal in the next N days”) beat the one that opens with the product name on all 4 profiles, so change_27 made task-first the template and Dana merged it. Replayed on the suite, POST /{namespace}/changes/{change}/review put the pick rate at 61 %; read back at +14d it is 58 %, n 120 picks, ■□□□ Too early to tell (Example, day 42).

Bring your own agent (folded) starts at You approve each one; a description it writes runs the same lint and sandbox test before any registry sees it.

npx startup-builder eval --scenario agent-registries-9 --agent "claude -p" --seat marketing-lead   # seated: 2.4 per 100 views, 0 lint blocks published, 9 of 9 installs passed · yours: 2.1, 0 blocks, 8 of 9

What it can do nextCopy Anchor Link

When it hands to youCopy Anchor Link

Failing first calls go to Product Lead · your agent; every reviewer ask comes to you. What lands in each queue, as captured: one Brief per Role (brief is Not live yet, gaps #7).

[
  {
    "$id": "https://api.sb/leaseline.app/tasks/task_267",
    "$type": "Task",
    "assignee": "worker_8",
    "due": "2026-10-02T17:00:00Z",
    "on": "issue_52",
    "brief": {
      "what": "5 of 14 callers from one directory sent { lease_id } where the Tool takes { unit }",
      "chose": "fix-the-first-call",
      "why": "registry callers fail their first call",
      "calls": 5
    }
  },
  {
    "$id": "https://api.sb/leaseline.app/tasks/task_271",
    "$type": "Task",
    "assignee": "worker_1",
    "due": "2026-10-09T17:00:00Z",
    "on": "tool_3",
    "brief": {
      "who": "a registry reviewer",
      "said": "Please link a privacy policy from the listing.",
      "chose": "answer-the-reviewer",
      "why": "a registry asked for a change",
      "draft": "Added: leaseline.app/privacy, linked from the listing and the manifest."
    }
  }
]

Each receiver answers with POST /{namespace}/tasks/{task}/claim (Take this one), POST /{namespace}/tasks/{task}/return with a reason or POST /{namespace}/actions/{action}/redirect, each a one-call window. Once you claim a reviewer ask, submissions to that registry wait for you. A Task nobody claims by its due time goes up a Role through POST /{namespace}/tasks/{task}/escalate (to you), and person-promise fires.

On time: you, 1 of 1 reviewer asks answered by its due time, n 1, ■□□□ Too early to tell; Product Lead · your agent, 2 of 2 fixes within 1 day. Your queue: counted by kind: in weeks 1 to 3, 9 listing approvals; since, about 1 refresh a week and 1 reviewer ask a month.

The fixed line: POST /{namespace}/deals/{deal}/assign sets owner to the receiving person, the agent is never a Deal’s owner, and Meetings land on the owner’s calendar under the owner’s name.

When it won’t sendCopy Anchor Link

The captured refusal, then the Stop that fired as a Campaign.paused reason (Not live yet, gaps #2); the description that published instead is the second thread in Example.

{
  "type": "https://startupbuilder.dev/errors/refused",
  "title": "Refused",
  "status": 422,
  "detail": "The description for send_renewal_notice did not publish: it says the Tool is free to try.",
  "checks": [
    {
      "check": "side-effect-hold",
      "status": "fail",
      "doc": "/plays/agent-registries#check-side-effect-hold",
      "fix": "Say the key it needs on the listing and the Tool.",
      "then": "Tool.update"
    },
    {
      "check": "lint-block",
      "status": "pass"
    }
  ]
}

[
  {
    "reason": {
      "limit": "install",
      "observed": "the install string failed on one registry after it moved its config format",
      "bound": "0 failed installs",
      "window": "1 test"
    },
    "doc": "/plays/agent-registries#stop-install",
    "issue": "issue_49"
  }
]

Did it workCopy Anchor Link

No listing existed before day 0 and every registry got one, so no held-back half exists, and the basis is before-after (rule o): the columns read Now and Before, and nothing prints as Added. Each figure’s call is POST /{namespace}/metrics/query by node; every node above the leaf is Not live yet (gaps #4).

{
  "$id": "decisions/decision_476",
  "$type": "Decision",
  "choice": "refresh-it",
  "reason": "version 1.4 renamed draft_notice to draft_renewal_notice, and 4 listings still show the old name",
  "stakes": "low",
  "options": [
    "list-it",
    "refresh-it",
    "test-a-description",
    "fix-the-first-call",
    "answer-the-reviewer",
    "park-it"
  ],
  "by": "worker_6",
  "on": "tool_3",
  "expected": "all 9 listings match the manifest within 1d, and first calls on draft_renewal_notice hold",
  "reopensWhen": "a listing still differs from the manifest at the next weekly crawl"
}

The leaf counts first calls with a real input from a registry-tagged install, per 100 listing views, by registry; health probes with empty input are dropped (31 of them). A registry that reports no views prints views unreported and counts in the parent only.

How it moves it: each of the 9 registries carries the task-first description and a tagged install line, where before day 0 none carried anything of Leaseline’s.

Guardrails: before against after: time-to-first-call for registry callers, 2 min 40 s, and 0 listings flagged or delisted.

{
  "$id": "https://api.sb/leaseline.app/campaigns/agent-registries",
  "$type": "Campaign",
  "play": "play:agent-registries@1",
  "path": [
    "registry-install-rate",
    "new-leads",
    "new-arr",
    "net-new-arr",
    "arr"
  ],
  "experiment": {
    "metric": "registry-install-rate",
    "unit": "per-100",
    "heldBack": 0,
    "decision": null
  },
  "readout": {
    "value": null,
    "n": 588,
    "maturing": 0,
    "interval": null,
    "readability": "inadmissible",
    "basis": "before-after",
    "pooled": true
  },
  "ledger": {
    "spend": 0,
    "workerCost": 3,
    "hours": 1.4,
    "wonArr": 347,
    "projectedArr": 192,
    "paybackDays": 4
  }
}

Outside benchmark: 97 % of 856 MCP tool descriptions held at least one defect, and fixing them lifted task success by a median of 5.9 points (a 2026 study of public MCP servers).

Before it · After it. Before it: agent-legibility makes the first call work cold. After it: agent-to-agent sells to the agents that call, and developer-relations answers the developers who claim a key.

ExampleCopy Anchor Link

Leaseline ran it from 2026-08-24 to 2026-10-19 (56 days): 1 MCP server with 3 Tools, listed on 9 registries, 3 a week. 5 of them report views: 588 in all. Read every captured listing at ?threads=all; three follow the tables.

The MCP Registry listing, Aug 26 · “list_leases_due: lists leases due for renewal in the next N days, by unit. draft_renewal_notice: drafts the notice for one lease. send_renewal_notice: sends it; needs a claimed key.”¹ Install: npx leaseline-mcp --via mcp-registry.²

Aug 30 · a developer’s first call, list_leases_due { days: 60 }, returned 12 leases; on Sep 18 their client, Maple Court, took Pro (+$347 ARR).

- send_renewal_notice: send renewal notices to tenants. Free to try.
+ send_renewal_notice: sends the notice for one lease to its tenant; needs a claimed key.

A registry reviewer, 2026-10-08 10:12 · “Please link a privacy policy from the listing.” Answer the reviewer went to Dana (task_271, due Oct 9, 17:00) with a drafted answer; she published the link and replied at 14:05 the same day.³

¹ Every sentence maps to the Tools manifest. ² Each install string carries its registry’s tag. ³ A reviewer’s ask goes to a person, on every rung.

Your settingsCopy Anchor Link

---
registries: all   # every registry that takes MCP servers or skills
---
Say plainly that send_renewal_notice needs a claimed key.
At most 3 new registries a week.

$ npx startup-builder show plays/agent-registries
registries all · the key line · 3 a week    your file  plays/agent-registries.mdx:2,4,5 (Ask 2026-08-24)
family demand · process launches · worker role:marketing-lead, owns never · chooses 6 (2 rule, 4 read) · approval first-send · checks lint-block, side-effect-hold, listing-cap, approved-pages · stops 5    catalog  play:agent-registries@1

Every key from the catalogCopy Anchor Link

Compiled, not written: edit the Play record agent-registries and its Leaseline run, then regenerate. Every step runs from a sentence or a line of code. In your words: the CRM words you know, and where each lives here.