
This guide is part of Launch and list your MCP server, within The MCP Guide. Want help getting found? See MCP marketing.
To get your MCP server into Claude's connectors directory, you submit it through Anthropic's developer portal at claude.ai/directory/manage and choose MCP connector. You need a remote server on HTTPS, OAuth 2.0 if your tools act on a user's account, a title and a read-only or destructive annotation on every tool, a privacy policy, documentation and a populated test account. Anyone on a paid Claude plan can submit.
After you submit, Anthropic scans the server automatically and, by default, lists it as a Community connector. Some listings are escalated to a slower Verified review where a person tests each tool. There is no separate application for Verified.
Everything below comes from Anthropic's own pages, checked on 6 November 2026. Review rules change, so re-read the submission guide before you start. If you need the basics first, read what Claude connectors are.
Key Takeaways
- Submit at claude.ai/directory/manage from a Pro, Max, Ultra, Team or Enterprise account; Free accounts can't submit, and on Team and Enterprise an Owner submits.
- Only remote servers are accepted as connectors; local MCPB bundles are no longer accepted and must go inside a plugin instead.
- Every tool needs a
titleandreadOnlyHintordestructiveHint, a name of 64 characters or fewer, and a narrow, accurate description. - Reviewers reject catch-all tools that mix read and write methods, prompt-injection wording, and connectors that move money or generate images, video or audio with AI.
- New listings are Community by default; Anthropic decides on Verified escalation itself, and review time isn't fixed.
- Your listing slug is permanent, and you delist a connector by emailing mcp-review@anthropic.com.
Who can submit
Anthropic's Publish to the directory page says "Anyone on a paid Claude plan can submit, there's no partner program to apply to first." It sets three checks:
| Check | Rule (checked 6 November 2026) |
|---|---|
| Plan | Pro, Max, Ultra, Team or Enterprise. "Free accounts can't submit." |
| Role | Pro, Max and Ultra submit from their own account. On Team and Enterprise, an Owner submits. Enterprise Owners can grant a Directory permission through a custom role. |
| Organization | "The listing belongs to the organization you submit from." Submit from the organization that should own it long term. |
There is no fee for the listing in any page we opened. The page also says there is no separate Claude Marketplace submission: Marketplace listings come from the directory.
Connector or plugin?
The directory takes two kinds of listing: an MCP connector (one remote server) and a plugin bundle (a GitHub folder with skills, commands, agents and MCP server references).
Anthropic's advice for a product with a remote server is to make both submissions: "the server as an MCP connector, then the plugin bundle whose skills teach Claude how to use it." Submitting the server as a connector gives your organization its listing, its dashboard and the option to pair it with the plugin.
Local servers are the exception. The submission guide says "desktop extension listings in the directory are deprecated, and the directory no longer accepts local servers packaged as MCP Bundles (MCPB)." To distribute a local server, include it in a plugin. Our page on MCP desktop extensions explains MCPB.
The submission requirements
The submission guide lists five requirements every connector must meet:
- Security: "meet Anthropic's security standards."
- Tool annotations: "every tool includes a
titleand the applicablereadOnlyHintordestructiveHint." - Authentication: "use OAuth 2.0 for authenticated services."
- Privacy policy: a privacy policy URL is collected in the listing step. Local connectors have extra README and manifest rules.
- Documentation: "provide clear setup and usage instructions."
All listings must also follow the Anthropic Software Directory Policy and Directory Terms. The policy adds requirements worth reading in full. It asks for verified support contacts, "a standard testing account with sample data," at least three working example prompts, proof you own or control the endpoints and domains you connect to, and notice to Anthropic of material changes. For remote servers it requires "secure OAuth 2.0 with certificates from recognized authorities" when authentication is needed, and says servers should support Streamable HTTP, with SSE "acceptable for now but will be deprecated."
Authentication details that trip people up
Anthropic's authentication guide covers Claude's OAuth client. Key points:
- Register the redirect URI
https://claude.ai/api/mcp/auth_callbackfor claude.ai, Desktop, mobile and Cowork. - Accept a loopback redirect on any port for Claude Code, for both
localhostand127.0.0.1. - Support S256 PKCE, and answer unauthenticated requests with a
401that points to your protected resource metadata. - Respond within 10 seconds on discovery, registration and token endpoints, and 30 seconds on refresh.
- Machine-to-machine
client_credentialsgrants are not supported.
For high-traffic servers, Anthropic says to prefer a Client ID Metadata Document (CIMD) or Anthropic-held client credentials over Dynamic Client Registration, because DCR registers a new client on every fresh connection. Our guide to MCP OAuth redirect errors helps when sign-in fails.
What reviewers reject
Anthropic's connector review criteria are organized around "the most common rejection reasons." The main ones:
| Area | What fails review |
|---|---|
| Tool split | One tool that accepts safe methods (GET, HEAD, OPTIONS) and unsafe ones (POST, PUT, PATCH, DELETE). "Don't ship a catch-all api_request tool with a method parameter." |
| Custom query tools | A freeform endpoint or query tool whose description doesn't link to or name the target API |
| Annotations | Missing title, readOnlyHint or destructiveHint |
| Names | Tool names over 64 characters |
| Descriptions | Wording that tells Claude to call tools the user didn't request, pulls instructions from outside sources, hides instructions, or promotes products |
| Errors | Generic errors such as "Internal Server Error" with no detail |
| Data | Collecting conversation data beyond the tool's need, or querying Claude's memory, chat history or user files |
| API ownership | Calling APIs that aren't yours or that you don't legitimately proxy; "The MCP server domain should match your service." |
| Use cases | Transferring "money, cryptocurrency, or other financial assets," or generating "images, video, or audio through AI models" |
Design tools that produce diagrams, charts or UI mockups are allowed. Annotations also matter at runtime: the review page says read-only tools can run without per-call confirmation, and "destructive tools always prompt." Our guide to MCP tool annotations shows how to set them.
The 10 steps in the developer portal
The submission guide walks through each step of the portal. Your progress saves in your browser as you go.
- Connection. Paste your
https://server URL or pick a custom connector you already added. If users connect to different URLs, choose Multiple URLs or URL pattern. Those listings can be published only once they have the Verified label. - Tools. Tools, prompts and resources sync from your server, grouped as read-only or write. Fix any flagged missing titles or annotations on the server.
- Listing. Server name up to 100 characters, one-liner up to 200, description up to 2,000, one to five categories, documentation URL, privacy policy URL, support contact, icon, and a URL slug. "The slug is permanent once published."
- Use cases. Primary use cases, what users need before connecting, and whether it reads, writes or both.
- Company. Company name, website and a primary contact.
- Authentication. Whether sign-in is required, which methods, and optionally your OAuth client ID and secret for Anthropic to hold.
- Data handling. Whether the API is yours, proxied with permission, or a third party's, and whether you handle personal health data or sponsored content.
- Test & launch. Step-by-step reviewer access with credentials for a fully populated account. You confirm you ran every tool in MCP Inspector or as a custom connector.
- Compliance. Seven required acknowledgments: directory guidelines, first-party API usage, financial transactions, AI media generation, prompt injection, conversation data collection and public documentation.
- Review and submit. A final check. Quality warnings, such as very short answers, are shared with the review team.
For an MCP App (a server that shows interactive UI), add 3 to 5 PNG carousel screenshots at least 1000px wide, cropped to the app response. Video and GIFs are not accepted.
After you submit
Anthropic's submission status page lists the connector statuses: Draft, In review, Changes requested, Not approved, Approved and Published. Not every submission passes through each one. Approved is not live yet; someone who can edit the listing must select Publish.
On labels, Anthropic's verification page says listings "flagged as highly useful to Claude users" may be escalated to Verified review, which is "higher touch and slower." That escalation "is assessed automatically," and you don't need to act. Review time isn't fixed. For escalations, email mcp-review@anthropic.com.
After publication, Anthropic's after publishing page says you don't resubmit to change tools. You deploy changes to your server. Tool names shown on the listing are listing details, so update them with a listing edit that a reviewer approves. Your listing URL is https://claude.ai/directory/connectors/SLUG, and the slug can't change. To remove a connector, email mcp-review@anthropic.com. A published connector also gets a dashboard with server health and usage by tool, per the publish page.
Prep checklist before you open the portal
Use this list to get ready. It combines Anthropic's checklist with the review criteria above.
- Remote server on a public
https://URL, Streamable HTTP preferred - OAuth 2.0 with PKCE S256, the claude.ai callback URL and a port-agnostic loopback redirect registered
- CIMD or DCR supported, or an OAuth client ready to hand to Anthropic
- Every tool has a
title, the right hint, a name of 64 characters or fewer and a narrow description - No catch-all read/write tool; write actions split by create, update and delete where possible
- Clear error messages for invalid input; responses sized to the task
- Every tool run in MCP Inspector and as a custom connector in Claude
- Public documentation page live by your publish date, with at least three example prompts
- Privacy policy URL covering collection, use, sharing, retention and contact
- Support contact and company details
- Reviewer test account with realistic sample data
- Icon, listing text within the character limits, a slug you will keep forever
- For MCP Apps: 3 to 5 carousel screenshots and any allowed link origins you own
Our MCP server launch checklist covers the wider launch, including the open MCP Registry, which Anthropic says is separate from its directory.
Next step
eSEOspace's Claude connector development team builds remote MCP servers that work as Claude custom connectors and helps prepare directory submissions. Approval is Anthropic's decision, so we don't promise outcomes or timelines. Once you're listed, our MCP marketing service handles docs pages, launch content and discovery in AI search.
More in the MCP Guide
Latest research: State of MCP: October 2026
Frequently asked questions
Does it cost anything to list a connector in Claude's directory?
The Anthropic pages we opened on 6 November 2026 list no listing fee. You do need a paid Claude plan to submit, and Free accounts can't submit.
How long does Anthropic's review take?
Anthropic says review time isn't fixed. Most submissions are scanned automatically and listed as Community connectors. Some get a review from a person, and those times "vary with queue volume."
How do I get the Verified checkmark?
You can't apply for it. Anthropic escalates listings it flags as highly useful to a Verified review, in which reviewers run each tool. That decision is assessed automatically.
Can I list a server that uses an API key instead of OAuth?
The submission guide requires OAuth 2.0 for authenticated services. Anthropic offers static request headers in beta for a limited set of organizations, and a no-authentication option for public data. If each user needs their own account, use OAuth.
My server URL is different for each customer. Can it be listed?
Yes. Choose Multiple URLs or URL pattern in the Connection step. Anthropic says such a listing "can be published only once it has the Verified label," and a URL pattern can't use Anthropic-held client credentials.
Can I submit a local .mcpb server?
Not as a connector. Anthropic says the directory "no longer accepts local servers packaged as MCP Bundles (MCPB)." Include the local server in a plugin bundle instead.











