The directories are filling up
Node8's August 2026 census counted 1,253 Claude connectors and 2,305 ChatGPT listings. Only 392 of 3,150 products were on both.
Source: Third-party census: Node8, August 2026
Last verified
Agent distribution engineering for B2B SaaS
Get your SaaS listed, working and chosen inside Claude, ChatGPT and Cursor.
Invokary turns your existing API into an MCP connector built to Anthropic's and OpenAI's review rules, submits it from your own accounts, and measures usage after launch. Fixed price. Your code, your domain.
Where you and two competitors stand in Claude, ChatGPT, Cursor and the MCP Registry, with 3 quick wins. A private page in 2 business days. No call needed.
- 8 to 20 tools built around real tasks, not your endpoint list
- Evals in CI before every submission
- Your repo, your domain, your listing
Our own MCP server is listed in Claude's Connectors Directory, ChatGPT's Plugins Directory and the official MCP Registry.See how we built it
We build and submit for: Claude Connectors Directory · ChatGPT Plugins Directory · Cursor Marketplace · Official MCP Registry.
Invokary is an independent studio. We are not affiliated with or endorsed by Anthropic, OpenAI or Anysphere.
Your customers are asking agents to use your product. Is your product there?
The protocol just changed
MCP 2026-07-28 removed sessions and the initialize handshake, made servers implement server/discover, and deprecated Dynamic Client Registration in favor of Client ID Metadata Documents. DCR still works for backward compatibility, for now. Servers built last year need work.
Source: MCP blog, July 28, 2026: The 2026-07-28 specification
Source: MCP specification 2026-07-28: Key changes
Last verified
Listed is not the same as used
Agents choose tools from names, descriptions and annotations. Claude now shows plugin publishers installs, listing views and the searches that lead to them. You can finally measure whether you are chosen.
Source: Anthropic, September 25, 2026
Last verified
What Agent Distribution Engineering means.
Agent Distribution Engineering is the work of making an existing SaaS product listed, working and chosen inside AI agents: tool design, auth, evals, directory submission and measurement after launch. We don't build agents. We make your product usable by the agents your customers already use.
- Not generic MCP development: a server is one deliverable; we're accountable for the listing, the review and whether agents pick your tools.
- Not a replacement for AI coding tools: they write code; we own tool design, review rules, evals and submission.
- Not an OpenAPI-to-MCP generator: generators mirror endpoints; we design 8 to 20 tools around real jobs.
Listed, working and chosen, then growing. We measure each one.
01
Listed
A live public listing on every surface you contract, published from your organization.
Measured by listing URLs and tier (Community or Verified).
02
Working
Every tool succeeds on valid input and auth works first time.
Measured by eval pass rate of 90% or more, tool error rate under 2%, uptime of 99.5% or more.
03
Chosen
The agent picks the right tool, and people find your listing.
Measured by tool-selection accuracy of 85% or more, plus listing views and search queries where the platform reports them.
04
Growing
Adoption rises month over month.
Measured by installs by surface and version, weekly tool calls and connected accounts over 90 days.
From your OpenAPI spec to a reviewed connector.
We do not generate one tool per endpoint. We map your API to 8 to 20 tools shaped around real jobs, because models get unreliable with large tool lists. Every tool ships with a title, the right read-only and destructive hints, and a description written for the model, not at it.
01API
We start from the API you already have, and from your OpenAPI spec if there is one.
02Tool design
8 to 20 tools shaped around real jobs, each with a title, read-only and destructive hints, and a description written for the model.
03MCP server
A remote MCP server on the 2026-07-28 spec, in your repo and on your domain.
04OAuth
OAuth 2.1 with Client ID Metadata Documents, plus a Dynamic Client Registration fallback.
05Evals
Five positive and three negative prompts per platform, run in CI.
06Submission
A submission package for each platform, submitted from your accounts.
07Discovery
Listings in the directories you choose, with names and descriptions that help agents pick your tools.
08Usage
Installs by surface and version, weekly tool calls and connected accounts after launch.
Built on the current MCP spec (2026-07-28), with OAuth 2.1 and Client ID Metadata Documents, evals in CI and code in your repo.
Why not have Claude Code or Cursor write it?
You can, and for an internal tool you should. A listed connector is a different job: each platform reviews it against its own rules, real customers sign in through OAuth, and a model has to pick your tools from everything in its context. An AI coding assistant can write a working server. The rest is what we sell: choosing 8 to 20 job-shaped tools, auth that passes review, evals that show the right tool gets picked, submission from your accounts, and fixes when reviewers or the spec change. For example, OpenAI's reviewers need test accounts that work without MFA, and projects with EU data residency can't submit plugins.
| In-house with AI coding tools | OpenAPI-to-MCP generator | Freelance MCP developer | Invokary | |
|---|---|---|---|---|
| Best when | Internal tools, or you have MCP and review experience in-house | You want a quick server from an existing spec | You have a defined spec and will run review yourself | You want a reviewed, listed connector at a fixed price |
| Tool design | You decide | Mirrors your endpoints by default | Depends on the developer | 8 to 20 job-shaped tools with annotations, approved by you |
| Auth that passes review | You design and test it | Varies by product | Depends on the contract | OAuth 2.1 with Client ID Metadata Documents, DCR fallback |
| Evals | You build them | Varies by product | Depends on the contract | In CI: 5 positive and 3 negative prompts per platform |
| Submission and review responses | Your team | Your team | Depends on the contract | From your accounts; 2 rejection cycles included |
| After launch | Your team | Your team | Depends on the contract | 30 days of monitoring, then Care or a handover |
| Price certainty | Engineering time | Tool pricing | Hourly or fixed | Fixed price, published acceptance criteria |
Not a hosting platform. Not a generic agency.
- Your code, your domain, your listing. We never host your production server.
- Fixed prices and published acceptance criteria.
- Built for the current spec, with a written plan for the next deprecation.
- We submit from your accounts, so the listing is yours forever.
What we can show you today.
We are a new studio, so we will not show you logos we have not earned. Here is what we can show today: the requirements each platform's reviewers check, with sources and check dates, a dated log of platform changes, and our own open-source MCP server, listed in Claude's Connectors Directory, ChatGPT's Plugins Directory and the official MCP Registry.
Mohd. Ashhar, founder.
Full stack AI engineer who founded two AI products, Clauseium and ApplyForge, and builds LLM features for Ascendo AI, a client.
Who you work withFounding client program.
Our first five Launch Sprints are $7,500 instead of $9,500, in exchange for a public case study with your approval.
Reference implementation (not a client)
Our own MCP server: six read-only tools with full annotations, checked in CI against the ChatGPT submission's test cases, and listed in Claude's Connectors Directory, ChatGPT's Plugins Directory and the official MCP Registry.
View reference workFixed scopes. Published prices.
- Agent Distribution Audit
$750
5 business days
Your directory presence, competitor map, tool map, auth plan and a fixed quote. Credited to your build.
- Connector Launch Sprint
From $9,500 ($7,500 founding)
3 to 4 weeks plus platform review
A remote MCP server in your repo, OAuth, evals in CI, and submissions to the platforms you choose.
- Connector Upgrade
$2,500 to $5,000
1 to 2 weeks
Move to 2026-07-28, add Client ID Metadata Documents, fix annotations, respond to a rejection.
- Connector Trust Pack
$1,500 to $3,000
1 week
Threat model, scope matrix and a reviewer-ready security statement.
- Connector Care
From $750 per month
3-month minimum
Monitoring, monthly evals and spec and policy watch, with a Listed, Working, Chosen, Growing scorecard.
Get a price band in two minutes.
Answer seven quick questions about your API, auth and platforms, starting with three here. You get a scope, timeline and price band. It is an estimate, and the Audit turns it into a fixed quote.
What happens after you book.
Day 0
15-minute fit call with Mohd. Ashhar. We tell you on the call if you are not a fit.
Within 1 business day
We send the invoice. The Audit starts when payment clears.
Days 1 to 5
Audit. You get a written report and a fixed quote.
Weeks 1 to 4
Build. You get weekly demos, and code lands in your repo from day one.
Submission
We prepare everything and submit from your accounts. Review time is set by each platform, and we respond to feedback within 2 business days.
Launch and 30 days
You publish. We watch errors and usage, then hand over or move to Care.
We track what each reviewer actually checks.
Claude
Claude Connectors Directory
Claude requirementsSingle MCP connector or GitHub-hosted plugin bundle, automated safety scan, connectors listed as Community by default, usage analytics after launch.
Last verified
ChatGPT
ChatGPT Plugins Directory
ChatGPT requirementsDomain verification token, verified identity, five positive and three negative test cases, tool annotations, release notes and a demo recording.
Last verified
Cursor
Cursor Marketplace
Cursor requirementsPublic Git repository with a plugin manifest, manual review, permissive licenses for open-source components.
Last verified
MCP Registry
Official MCP Registry
MCP Registry requirementsA server.json listing your public endpoint, under a namespace you authenticate. The registry is in preview.
Last verified
Questions CTOs ask us
Can you guarantee approval?
No. Anthropic, OpenAI and Cursor decide. We build to their published requirements, fix what reviewers flag, and include two rejection cycles.
Who owns the code?
You do, on payment. It lives in your repository from the first commit.
Do you host the server?
No. It runs on your infrastructure or your Cloudflare, Vercel or AWS account.
What do you need from us?
An engineering contact for about 2 hours a week, OAuth admin access, a reviewer test account with realistic data, DNS access and a Claude Pro, Max, Team or Enterprise plan (on Team and Enterprise, an Owner submits).
We already have an MCP server.
We audit it, migrate it to 2026-07-28 and fix what blocks listing. That is the Upgrade.
How is this different from an OpenAPI to MCP generator?
Generators mirror endpoints. Reviewers and models reward a small set of well-described tools with correct annotations and real auth. See how the options compare.
Where are you based?
We are an engineering studio based in India, working with SaaS teams in the US, UK and EU. We work 09:00 to 18:00 IST, which covers UK and EU mornings. For US teams we hold calls early morning Eastern time, or later by arrangement. We reply within 1 business day.
How do we pay?
By invoice in USD, paid by bank transfer. The Audit is paid upfront; a Launch Sprint is half at kickoff and half at submission.