Claude Code can work on your product code while customer questions live in a separate support dashboard. Connecting the two makes a useful workflow possible: inspect what visitors actually asked, identify a specific gap, then prepare a change with evidence attached.
This guide connects Claude Code to one Namiru agent through MCP. You need Namiru Pro, access to the agent's Settings page, and a Claude Code installation that supports HTTP MCP servers. Start with a read-only key.
Create a scoped key in Namiru
Open your agent's Settings and expand AI workspace connection (MCP). Give the key a recognizable name such as “Claude support review” and choose an expiry. Leave Allow settings changes off. Copy the key immediately; Namiru does not show it again.
Use the URL displayed in that panel. The production endpoint is https://api.namiru.ai/api/mcp. If you are working in an isolated test environment, copy its test URL instead.
Add the HTTP connection
Run the following in your own terminal, replacing YOUR_KEY locally:
claude mcp add --transport http namiru https://api.namiru.ai/api/mcp --header "Authorization: Bearer YOUR_KEY"
Keep the resulting credential-bearing configuration out of shared repositories. A command containing a real key may remain in shell history. Use your team's credential handling conventions and revoke a key if it is exposed.
Open Claude Code and use /mcp to inspect the connection. The command format is documented in Claude Code's official MCP guide. Namiru's setup uses a bearer key, not an OAuth login flow.
Ask a question with a verifiable answer
Try this prompt:
Review the last seven days of conversations for repeated questions about opening hours. Read the matching transcripts, distinguish unanswered questions from successful answers, and cite the conversation IDs. Do not modify settings or contact visitors.
The expected tool sequence is list_conversations, followed by get_conversation for relevant IDs. Make sure the client follows nextOffset on paginated results. A search for one phrase can miss synonyms, so begin with a date range and use more than one search when necessary.
The output should separate observations from proposals. “Three visitors asked whether Saturday appointments are available” is an observation only when three source conversations support it. “Make Saturday availability more prominent” is a proposal. A small sample is not proof that all customers have the same problem.
Review a settings change
If you want Claude to apply changes, create a separate key with settings write access. Ask it to read get_settings and get_settings_schema, explain the exact proposed values, and apply only the intended change with update_settings.
For example, it can update agent instructions to refer visitors to the configured booking flow. It should not invent opening hours from incomplete conversations. Keep the business schedule authoritative. Nested objects are replacements, so preserve existing fields when updating an object such as the agent configuration.
Credential values are not returned by settings reads. Replacing a webhook or tool configuration may require supplying its credentials again. Google authorization, billing and account security remain dashboard tasks.
Troubleshoot connection failures
A 401 response means the key is missing, invalid, expired or revoked. A 403 can indicate that effective Pro access is unavailable or the request has an untrusted browser origin. A tool validation error means the requested input did not match its schema; inspect the schema and correct the field rather than retrying the same payload.
For an unknown conversation, confirm you used the conversation's id rather than sessionId, and that the key belongs to that agent. Keys cannot cross agent boundaries.
Namiru MCP is included in the Pro package, priced at €59 monthly or €588 annually. For the full tool and permission model, read the Namiru MCP overview.
