How to get listed in the Claude connector directory

It is the list of connectors Claude users browse under Settings, Connectors, where each entry is a remote MCP server Anthropic has reviewed. A connected server works across Claude on the web and mobile, Claude Desktop, Claude Code and the Claude API.

Search interest is small but well timed, about 110 searches a month for "claude connector directory" and 40 for "claude connector submission", from people who are about to connect or submit a server.

Because Claude then uses your product to answer its users, and no outside page can make that happen. For an MCP product it's the strongest agent reach available, alongside ChatGPT's apps directory.

A link from a listing page, if one shows up, is a bonus. What you want is for Claude to pick your tools when someone asks for the job they do.

The form asks for listing basics, how Claude connects, up to three use cases, prerequisites, read and write capabilities, a policy checklist and company details. Tool and prompt names are read from your server, not typed in.

The policy checklist asks you to confirm that tool descriptions hold no instructions about model behaviour or other tools, that the server collects no conversation data it doesn't need, and that public documentation is live by launch. People break the first one by accident, by telling the model in a description which tool to call next.

Write each one as a task a person would ask Claude to do, with a prompt you have run against your own server. Reviewers use them to test the listing, and they aren't shown publicly, so be concrete.

These are the prompts we submitted:

• "Audit https://example.com/pricing and tell me what to fix first."
• "Which AI crawlers may read https://example.com?"
• "Draft an llms.txt for example.com."

Run every prompt in Claude with your connector added and check that Claude calls the tool you expect. If it picks another tool or answers from memory, fix that tool's description, not the prompt. Our guide to MCP tool descriptions covers what to write.

Check that it answers a plain MCP handshake without credentials, if you chose no sign-in, and that every tool carries honest annotations. The portal labels each tool "Read-only" or "Writes" straight from its readOnlyHint.

curl -s https://mcp.workpods.ai/mcp -X POST \
  -H 'content-type: application/json' \
  -H 'accept: application/json, text/event-stream' \
  -d '{"jsonrpc":"2.0","id":1,"method":"initialize","params":{"protocolVersion":"2025-06-18","capabilities":{},"clientInfo":{"name":"check","version":"1"}}}'

The reply should carry your server's name, title, version, description and website. A tool that sends something, starts a job or saves something is not read-only, so mark it readOnlyHint: false. Claude also takes each tool's display name from annotations.title, so set that too. The fields directories read from the handshake are in our MCP server metadata guide.

Deploy the change, open the submission in the full editor, click Sync from server, then Save and update submission. Reviewers see what you save, so you don't resubmit.

claude.ai/directory/manage → your submission
  Edit → Open the full editor
  Sync from server             reads your live server, up to a minute
  Save and update submission
  Tools tab                    check the date and each Read-only / Writes label

The portal keeps a dated copy of your tools and reviewers read that copy. We synced after renaming our tools, after adding a description, website and icon to the handshake, and again when we added a tool that saves a report only on request. If the date on the Tools tab hasn't moved after saving, the save didn't go through.

Open Settings, then Connectors, choose Add custom connector and paste the server URL. Anyone can do this before your listing is approved, and in Claude Code it takes one command.

claude mcp add --transport http agent-ready https://mcp.workpods.ai/mcp

Document both ways on your own site, since the checklist asks for public documentation and early users need somewhere to start. Ours are on the Agent Ready tool page, and the next step for most servers is the MCP Registry. Every channel is ranked in our MCP directory guide.

To check what agents see on your site and server before you submit, run Agent Ready. If you'd rather we prepare and submit the listing, that's part of our agent platform work.