---
title: "Connect a remote MCP server to Claude Code | FidelicAI"
description: "Connect an approved remote MCP server to Claude Code, check project access, and verify its tools. The first FidelicAI MCP beta is free during testing; the connection guide supplies the published MCP address."
canonical: "https://fidelic.ai/integrations/claude/connect-mcp"
lastUpdated: "2026-10-13"
---

# Connect an approved MCP server to Claude Code

Claude Code can use an approved remote MCP server. Confirm the actual publisher, endpoint, account requirements and permissions before adding it. The first FidelicAI MCP beta is free during testing; the connection guide supplies the published MCP address.

**Updated:** October 13, 2026.

## Cookbook connection and testing access

[Website migration guidance](https://fidelic.ai/work/website-migration) is public. The [local discovery manual](https://fidelic.ai/catalog/local-search-and-ai-discoverability) is publicly available for free use during testing. Use the published MCP address in the connection guide. Earlier role-hiring connector instructions do not establish access to this manual.

## Add and verify the approved connection

### 1. Choose the project and verify the publisher

Open the intended project. Read the publisher’s current endpoint, authentication instructions, tool list and terms. Do not paste credentials into a chat or repository.

**Done when:** the owner has approved the publisher, project scope and required access.

### 2. Add the HTTP server

Replace the example URL with the publisher’s approved address. Project scope can be shared through .mcp.json; review it before committing. The example is not a FidelicAI endpoint.

```sh
claude mcp add --transport http --scope project approved-expertise https://your-approved-server.example/mcp
```

**Done when:** the project records the intended server name and actual approved endpoint.

### 3. Inspect tools and authentication

Start Claude Code and use /mcp to inspect status and authenticate where required. Confirm each tool’s purpose and requested access before use.

**Done when:** the connection status and permitted tools match the publisher’s current documentation.

### 4. Test one bounded read

Ask for public guidance on a defined job. Verify the returned source, scope, version and limitations before using it to change a business system.

**Done when:** the result identifies its source and an observable next step without claiming unavailable tools or completed changes.

## Read public methods without a server

[FidelicAI’s public reading index](https://fidelic.ai/for-ai) links to Markdown instructions and the catalog. Reading a public file is separate from installing a persistent skill or connecting a paid server.

## Sources

- [Anthropic: Connect Claude Code to tools via MCP](https://code.claude.com/docs/en/mcp)

- [FidelicAI connection status](https://fidelic.ai/connect/claude)

## Continue with your own Claude

The [FidelicAI Claude guide](https://fidelic.ai/integrations/claude) states the current public-skill and planned paid-access status.

- [Use Claude Code in Slack for small business](https://fidelic.ai/integrations/claude/claude-code-in-slack)

- [Get business expertise from a Slack thread](https://fidelic.ai/integrations/claude/hire-ai-agent-from-slack)

- [Prepare a podcast task with your own Claude](https://fidelic.ai/integrations/claude/prepare-sadie)

- [Keep expertise requests separate from account access](https://fidelic.ai/integrations/claude/security-and-permissions)

## Citation

FidelicAI. “Connect an approved MCP server to Claude Code.” https://fidelic.ai/integrations/claude/connect-mcp
