submit-to-claude
Skill1 starsChanged 9 days agoMITView on GitHub
Walk through listing a remote MCP server in Claude's connector directory at claude.ai/directory/manage - testing it first as a custom connector, letting Anthropic's servers through bot protection, the portal's ten steps (connection, tools, listing copy that fits the directory card, use cases, company, authentication, data handling, test account, compliance, review), card link permissions and screenshots, and pairing an optional Claude plugin bundle. Use when someone wants their MCP server, app or ChatGPT plugin in Claude's directory, is filling in the Claude connector portal, needs listing copy or categories, or their connector fails to link in claude.ai.
In this repository8 other files it gives its agents.
- AGENTS.mdAGENTS.md
- mcp-app-cardsSkill
- mcp-server-for-pluginsSkill
- plugin-oauthSkill
- plugin-packageSkill
- plugin-prerequisitesSkill
- publish-mcp-serverSkill
- submit-to-chatgptSkill
The file
We split this skill into the rules it gives an agent
5 rules. About 0 are common, 3 are distinctive, and 2 are very specific to this project.
≈0
Common · estimate
Many files like this have a rule of this kind.
≈3
Distinctive · estimate
This author's own choices. Worth a look.
2
Very specific
Tied to their project's names, files or tools. Probably skip.
- Common
- Distinctive
- Very specific
- Very specific: line 22
If linking bounces you back to the sign-in page or Allow vanishes, see the debugging list in plugin-oauth (another open tab signing you …
- Very specific: line 24
If you use bot protection, let Anthropic's published egress ranges through for /mcp and the OAuth endpoints.
- Distinctive: line 30
Tools: read from your server automatically.
- Distinctive: line 54
Every claim in the description must be something a tool returns today.
- Distinctive: line 58
After submitting, keep the generated portal text somewhere private (a private repo or your notes) so a later edit starts from what was …
≈ “Common” is an estimate: at least 1 in 5 of the 364,980 skill files we read have a rule of the same kind, not the same words. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.
Files
Submitting to Claude's connector directory
Submit from an account on a paid Claude plan. Anthropic's review checks the listing against
what the tools do and runs your test steps with the reviewer account. Every field, in the
portal's order: references/portal.md.
0. Test as a custom connector first
claude.ai → Settings → Connectors → Add custom connector → paste the MCP URL exactly
(https://…/mcp, no trailing character) → sign in on your page → Allow. Then in a new chat with
the connector on, run each use case and check the cards draw.
If linking bounces you back to the sign-in page or Allow vanishes, see the debugging list in
plugin-oauth (another open tab signing you out, bot protection, a stale tab after a deploy).
If you use bot protection, let Anthropic's published egress ranges through for /mcp and the
OAuth endpoints.
1. The ten steps (claude.ai/directory/manage → Submit new → MCP connector)
- Connection: the universal MCP URL.
- Tools: read from your server automatically; each needs a title and all four hints.
- Listing: name; one-liner (limit 200, but the directory card shows about 90 characters, so put the whole hook in the first 90); categories; documentation URL; privacy URL; support; icon; URL slug (permanent once published); description (≤2,000).
- Use cases: at least three, each a title, a prompt, and what Claude does with it; also "what someone needs before connecting" and whether the connector reads, writes or both.
- Company: name and website; a contact for review updates.
- Authentication: OAuth. Claude registers by Client ID Metadata Document (or DCR), so
there's no client secret to share. Its return address is
https://claude.ai/api/mcp/auth_callback. - Data handling: whose API it is, health data, sponsored content, and a plain statement of what you read and keep.
- Test & launch: the reviewer account's email (the password goes only in the credentials field) and numbered steps a stranger can follow; confirm you tested every tool in Claude.
- Compliance: seven acknowledgements; read each.
- Review and submit.
Cards that open links need those origins listed as allowed link URIs (scheme and host only, your own domains). Screenshots: 3–5 PNGs at least 1000 px wide, cropped to the card, with the prompt given separately.
2. Copy that passes review
- Lead the one-liner with what the user gets, not the technology.
- Every claim in the description must be something a tool returns today. If the listing promises a feature, check the connector actually exposes it, not just your website.
- Numbers counted on real data. Wording like "from the official source wherever one is published" when some data is unofficial.
- After submitting, keep the generated portal text somewhere private (a private repo or your notes) so a later edit starts from what was submitted. It holds the reviewer's email, so keep it out of a public repo, and the password never goes in it.
3. The optional plugin bundle
A Claude plugin (public GitHub repo) can bundle a skill that teaches Claude to use your tools
well, with .mcp.json pointing at the same MCP URL. Submit it separately and pair it with the
connector; see plugin-package for the files and the directory's checklist (README ≥40 words,
LICENSE, no symlinks or secrets).
Does it work?
Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.
No reviews yet
People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.
- It worked
- Partly
- It didn't
Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.
Your trail
Files you open will collect here, so you can find them again.
Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.
Now do this for your own.
Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.

