Help & Documentation
Connect AI tools to your DashHarbor website
Find your native EmDash MCP endpoint, create a scoped token and connect a compatible AI tool, with a tested Codex CLI example.
Connect a compatible AI assistant or development tool directly to your website using native EmDash MCP. You create the access token and choose what the tool can do.
New to MCP? Start with Understanding MCP and DashHarbor. To end an existing connection, follow Disconnect AI tools from your DashHarbor website.
Before you connect
Your AI client needs remote MCP, Streamable HTTP transport, compatible bearer-token authentication and a secure way to supply the token. Configuration differs between applications; the Codex commands later in this guide apply only to Codex CLI.
Check the compatibility information for your installed website release. Compatibility unverified means the installation is not covered by the current verified evidence. Contact support to confirm an unverified installation before using supported setup instructions. Do not treat the current theme catalogue or a visible endpoint as proof that your installed website has been tested.
The verified native personal-token checks cover EmDash 1.1.0 installations using Marketing 1.0.3, Blog 1.0.2 and Portfolio 1.0.3. These are tested release combinations, not separate MCP implementations. Codex CLI 0.159.3 passed staging checks and an earlier production read-only pilot on Marketing 1.0.1. Exact-release production read/write checks used the native MCP SDK; they do not establish complete production Codex acceptance on every release.
Other clients, EmDash releases outside that evidence, direct custom-domain MCP and OAuth/device sign-in remain unverified. This guide does not promise ChatGPT or Claude compatibility. More client examples can be added after testing.
1. Find the right website and endpoint
Sign in to DashHarbor and open the website you want the AI tool to access. Check its name and permanent hostname, especially if you manage several websites.
Where your dashboard shows AI & MCP connections, check the environment label and choose Copy endpoint. That section is currently available in the DashHarbor application staging dashboard; the production dashboard may still use the earlier layout.
On a dashboard without that section, choose Edit website. Keep the exact HTTPS hostname used by the native EmDash editor, and replace /_emdash/admin and everything after it with /_emdash/api/mcp.
The endpoint has this structure:
https://[website-hostname]/_emdash/api/mcpFor example:
https://your-site.getdashharbor.com/_emdash/api/mcpUse the permanent DashHarbor website hostname. Do not substitute your public custom domain, the account dashboard, an internal Worker address or EmDash's documentation server. A custom domain for visitors does not automatically make it a supported MCP endpoint.
Keep production and staging connections separate. Use the address supplied for the intended installation; do not add or remove hostname prefixes yourself.
2. Open EmDash and create a personal token
Open your website's editor from DashHarbor to sign in. Then choose Settings → API Tokens → Create Token. Where shown, the dashboard's API token settings link uses that existing editor session.
Give the token a recognizable name, such as My AI tool - Production read-only, and choose an expiry. Use a separate named token for each tool, website and environment so you can revoke one connection independently.
Token management requires the appropriate native EmDash account permission. If you cannot see it, check that you opened the right website and signed in through DashHarbor. Do not grant an AI client Admin scope to solve a connection problem.
3. Choose the permissions you need
For website inspection, start with Content Read and Schema Read only.
- Content Read (
content:read): read existing content. - Schema Read (
schema:read): inspect collection structures and field definitions. - Content Write (
content:write): create, update and delete permitted content, potentially including publication. - Media Read (
media:read): read media information. - Media Write (
media:write): upload media and modify permitted media information.
Content Write is not draft-only access. An owner's token can allow publication and deletion. Your native EmDash role and content ownership still apply. Add write or media permissions only when the task requires them.
EmDash shows a new token once. Enter it only in your client's supported secure credential mechanism. The DashHarbor dashboard does not collect the token value, and you should never send it to support.
4. Configure your AI client
In your application's MCP connection settings, supply the copied endpoint and select its supported Streamable HTTP and bearer-token options. Give the connection a name that identifies the website and environment.
Use the application's documented secret store or supported environment-variable mechanism. Do not paste a token into an ordinary AI conversation, URL, repository or shared configuration file. A client that only offers a different transport or sign-in method may not connect to this implementation.
5. Verify with a read-only request
Ask the connected tool to list content collections, inspect their schemas and read an existing item without making changes. Confirm that the returned content belongs to the intended website.
A successful connection is not permission to publish. If you later grant write access, read the current item and revision, preserve unrelated fields, save a draft and review the result in EmDash before publishing. Reread the item if a stale-revision conflict is reported.
Codex connection example
This example is for Codex CLI in Bash, using the existing tested personal-token setup. These are not universal MCP commands. You need your own Codex access and a compatible installed website release.
Replace the example URL with your actual endpoint. The prompt reads the token without displaying its characters or placing its value in the command history.
export DASHHARBOR_MCP_URL='https://your-actual-editor-host/_emdash/api/mcp'
read -r -s -p 'DashHarbor token: ' DASHHARBOR_MCP_TOKEN
printf '\n'
export DASHHARBOR_MCP_TOKEN
codex mcp add dashharbor-production \
--url "$DASHHARBOR_MCP_URL" \
--bearer-token-env-var DASHHARBOR_MCP_TOKEN
codexCodex stores the environment variable's name in this connection's configuration. Start it from the same terminal so it inherits the token value. Another terminal does not automatically share that value. This bearer-token setup does not require codex mcp login.
Use /mcp to inspect the connection, then ask:
Using the dashharbor-production connection, list my content collections and read their schemas. Do not create, edit, publish or delete anything.
Confirm the website identity. When finished, revoke the token in EmDash first, verify a fresh request is rejected, then remove the local connection and clear the variables:
codex mcp remove dashharbor-production
unset DASHHARBOR_MCP_TOKEN DASHHARBOR_MCP_URLIf you used another connection name, use that name in the commands. See the full disconnection checklist. The official Codex MCP instructions explain the client's configuration options.
Troubleshooting
- Connection cannot be reached: check the exact HTTPS hostname and
/_emdash/api/mcppath. Confirm the client supports remote Streamable HTTP and bearer tokens. - Authentication fails: check token expiry, revocation and the website it was created for. For Codex, confirm the process inherited the named environment variable. Opening the browser editor alone does not authenticate an external AI client.
- A tool returns a permission error: compare its required scope with your token and native role. Add only the permission needed; do not switch to Admin access.
- Token settings asks you to sign in: open the editor from DashHarbor first, then return to Settings → API Tokens.
- Wrong content appears: stop using the connection and check its hostname and environment before allowing any changes.
- Compatibility is unverified: contact support with the website hostname, installed version if known and a redacted error. Do not send the token.
Keep access under your control
Keep token values out of chats, support tickets, URLs, logs, analytics and Git. Use a secret manager for longer-term storage and set an expiry suited to the task. Do not reuse production credentials for staging.
Revoking a token stops new authorized requests; it does not erase content an application already read. Removing a local connection alone does not revoke the token. Follow Disconnect AI tools from your DashHarbor website when access is no longer needed.