MCP servers, done for you
Your product, inside ChatGPT and Claude.
We build the MCP server, host it, and get it ready for the Claude Connectors and ChatGPT Apps review. Fixed price, ten business days, and you keep the code.
Not sure where you stand? Start with the free score.
The bar is real
The directories are open. The review is not a formality.
Anthropic's Connectors Directory and OpenAI's Apps want OAuth 2.0, a title and a read-only or destructive annotation on every tool, a privacy policy, docs and a working test account. A missing privacy policy is an immediate rejection, in Anthropic's own words.
Almost nobody is there yet
Of 690 major sites we scanned, 8% expose a single tool to an agent.
The median Agent Readiness Score across that set is 49 of 100. Being usable from inside a chat is still a way to be first in your category, not a way to catch up. See the leaderboard.
Fixed price, not a quote
Agencies quote $8,000 to $25,000 and four weeks. We quote $4,900 and ten business days.
The build runs on AI tooling under a person who ships MCP servers for our own products. The price is the price; it does not grow during the project.
Packages
Three things for sale. Pick the one that matches where you are.
Have an API and no MCP server: Build. Have a server that has not made it through review: Listing Sprint. Have it listed and want it to stay that way: Care.
Listing Sprint
$1,900 one-time
Your existing MCP server or API, made directory-ready.
5 business days
- Every tool gets a title and the right readOnlyHint or destructiveHint
- OAuth 2.0 flow checked against what the Claude and ChatGPT reviews expect
- Privacy policy and setup docs written for the listing
- Test account and reviewer instructions prepared
- Screenshots for MCP Apps, cropped to spec
- Fix list with code for everything the review would reject
- 30 minutes together in the submission portal
Build
$4,900 one-time
50% deposit to start, balance on delivery
An MCP server for your API or website, built, hosted and submitted.
10 business days
- Up to 8 tools from your REST, GraphQL or OpenAPI, or a read layer over your website
- Streamable HTTP transport on the current protocol version
- OAuth 2.0 with dynamic client registration where the directory requires it
- Staging URL and MCP Inspector access while we build
- Uptime and tools/list drift monitoring for the first 30 days
- The full Listing Sprint on top, so it goes into review the day it is done
- Code handed over in a repository you own
Care
$390 per month, cancel monthly
Hosting, monitoring and updates when the spec or the directories move.
ongoing
- Hosting on our server in Germany, or on your infrastructure
- Uptime checks and a drift alert when a tool or a description changes
- Protocol and SDK updates as the MCP spec moves
- Re-submission after a change that the directory has to see again
- Up to two hours of changes per month
- A monthly note on what agents did with your tools
Prices in US dollars, invoiced by FINAL MASTER LLC. EU businesses with a VAT ID are invoiced without US sales tax. Card or bank transfer.
How it works
From a domain to a listing in four steps.
- 01
Check
Run the free Agent Readiness Score on webmcp-tool.com. It says what agents can already do with your site, and what they cannot. No account.
- 02
Kickoff, thirty minutes
What your API does, which eight tools matter to a user talking to Claude or ChatGPT, who in your company owns the Claude Team or OpenAI account that submits.
- 03
Build
A staging URL within days. You add it to Claude as a custom connector and use it before anyone else does. MCP Inspector access, tool by tool.
- 04
Review
We prepare every field of the submission and sit with you while you click submit from your organisation. If the reviewers flag something, one fix round is included.
Who you are talking to
The people behind webmcp-tool.com and agenttracking.co.
mcp-tool.com is run by FINAL MASTER LLC. We built the Agent Readiness Score, a 23-check methodology that has scanned more than 800 sites, and Agent Tracking, an open-source dashboard of what AI agents do on a website. Both sites run their own MCP servers in production. That is the experience the fixed price rests on.
You talk to Gorden Wübbe, who runs the projects himself. No account manager in between.
Before and after
- Before: webmcp-tool.com measures whether agents can read, understand and operate your site. Free, no account.
- During: mcp-tool.com builds the server and gets it listed.
- After: agenttracking.co shows which agents come and what they do with your tools. Open source, one line of script.
Questions
The things people ask before they buy.
Who submits the connector to the directory?
You do, from your own organisation. Anthropic's portal lives inside Claude.ai under a Team or Enterprise organisation and only an Owner can submit; OpenAI submissions run through your own developer account. We prepare every field, every asset and the test account, and we sit with you for the thirty minutes it takes to click through. What we never do is submit under our name and hold your listing hostage.
Do we need a Claude Team plan?
For the Claude Connectors Directory, yes: submission is only available on Team and Enterprise. For ChatGPT Apps you need an OpenAI platform account. Neither is expensive next to the build, and we tell you on the kickoff call exactly which one applies.
Is acceptance into the directory guaranteed?
No, and anyone who guarantees it is guessing. The reviews are run by Anthropic and OpenAI, not by us. What we guarantee is that the server meets every published requirement when it goes in, that one round of reviewer feedback is fixed inside the package, and that you can use the connector in Claude and ChatGPT as a custom connector from day one, listed or not.
Who owns the code?
You do, on full payment. It is handed over in a repository under your account, with a licence that lets you do anything with it. We keep the right to reuse our own generic building blocks on other projects; nothing specific to your product goes anywhere.
Where is the server hosted?
On your infrastructure if you have one, otherwise on ours: a server in Germany that already runs our own MCP endpoints. Care keeps it there; without Care you take the repository and run it wherever you like, with a deployment script that works.
What if we do not have an API?
Then we build a read layer over what you have: a website, a catalogue, a knowledge base, a booking page. Read-only tools are the ones the directories accept most easily, and for many products they are all a user in a chat needs.
How is a fixed price possible when agencies quote by the hour?
Because the build runs on AI tooling with a person reviewing every tool, and because we have done it for our own products first. The scope is written down before we start: up to eight tools, one API, one OAuth flow. Anything beyond that is quoted separately, in writing, before it is built.
What do you need from us to start?
API documentation or a description of what the product does, a test account with data in it, someone who can answer questions within a day, and, for the submission, the person who owns your Claude or OpenAI organisation. The ten business days start when we have those.
Start with the score, or with a call.
Both take a minute. Neither needs an account.