MCP Server

What is MCP

MCP stands for Model Context Protocol. It is an open standard for connecting AI assistants and AI agents to external systems. Think of it as a common language: any MCP-compatible assistant can use it to find out what a system can do, then use that system in a structured, predictable way.

Inspectorio runs an MCP server. It gives an AI assistant a way to read and write your Inspectorio data directly, using plain-language requests instead of custom API code.

What the MCP server covers

The MCP server is enabled for all of Inspectorio's integrations. It is not limited to one part of your account. Whatever your API Key can already reach through direct API integration, your AI assistant can also reach through the MCP server.

You connect once. You do not need a separate connection, login, or key for each part of your account.

Why use the MCP server

The MCP server does not replace your existing integration. It is an additional, faster way in, and it fits different situations than a REST API, File Transfer, or Webhook integration does:

• Explore before you build. Ask an assistant to look something up or make a small change, before you invest time writing and maintaining a full API client.
• Give non-developers a way in. A teammate who cannot write code can still ask an assistant to check a record or make an update, in plain language, without waiting on engineering time.
• Speed up day-to-day support and troubleshooting. Instead of clicking through several screens or writing a one-off script, ask your assistant to find what you need and act on it.
• Cut down on small, repetitive integration work. For low-volume or occasional tasks, an assistant call can replace a small script you would otherwise have to write, test, and maintain over time.
• Keep pace as your AI tools change. As your organization adopts more AI assistants and agents, each one can reach Inspectorio the same way, without a new integration project for every tool.

For high-volume, scheduled, or event-driven work, keep using API Integration, File Transfer, or Webhooks. The MCP server is built for flexible, on-demand use, not for moving large volumes of data on a fixed schedule.

How it works

1. Your AI assistant connects to the Inspectorio MCP server over the internet, using a standard MCP transport.
2. You supply your existing Inspectorio API Key(s) when you connect. No separate MCP account or login is needed.
3. The assistant asks the server what it is allowed to do with your key(s), and lists those actions for you or uses them directly, depending on how you have set up your assistant.
4. Every action the assistant takes on your data goes through the same checks your API Key already goes through today. Inspectorio's platform, not the MCP server, is the final authority on whether an action is allowed.

Because of step 4, the MCP server cannot do anything your API Key could not already do through direct API integration. It changes how you reach your data, not what you are allowed to do with it.

Endpoints

Environment	URL
Pre Production	https://integration.pre.inspectorio.com/mcp
Production	https://integration.inspectorio.com/mcp

Test your setup against Pre Production first. Point your client at Production when you are ready to go live.

Authentication

The MCP server uses the same API Key(s) you already use for direct API integration. See Authentication if you need to create one.

Send your API Key(s) as request headers when your client connects, not inside an individual request to the assistant. Which header you send depends on what you want your assistant to reach:

• x-rsc-apikey — needed only for Responsible Sourcing & Compliance (RSC) actions.
• x-qrm-apikey — needed for everything else.

You only need to send the header(s) for the part(s) of your account you want the assistant to reach. A request that needs a header you did not send will be rejected.

Connect a client

The server speaks the Streamable HTTP transport. Any MCP client that supports it, and lets you set custom request headers, can connect.

Claude Code

Use the CLI:

claude mcp add --transport http inspectorio https://integration.inspectorio.com/mcp \
  --header "x-qrm-apikey: <your API Key>" \
  --header "x-rsc-apikey: <your API Key>"

Or add it directly to .mcp.json in your project:

{
  "mcpServers": {
    "inspectorio": {
      "type": "http",
      "url": "https://integration.inspectorio.com/mcp",
      "headers": {
        "x-qrm-apikey": "<your API Key>",
        "x-rsc-apikey": "<your API Key>"
      }
    }
  }
}

Only include the header(s) for the part(s) of your account you use. Run /mcp inside Claude Code, or claude mcp list from your terminal, to confirm the connection.

Claude Desktop / claude.ai

Go to Settings → Connectors → Add → Add custom connector, and paste in the endpoint URL.

This server authenticates with request headers, not OAuth. Header-based authentication for custom connectors is currently a beta feature: an organization administrator enters the credential once, when adding the connector, and it applies to everyone in the organization. Standard header names, such as authorization and x-api-key, are accepted without extra steps. Because this server uses x-qrm-apikey and x-rsc-apikey, your administrator may need approval for those specific header names before the connector can be saved this way. If that approval is not yet in place for your organization, Claude Code is the more direct path today.

Codex CLI

Add the server to your config.toml:

[mcp_servers.inspectorio]
url = "https://integration.inspectorio.com/mcp"
http_headers = { "x-qrm-apikey" = "<your API Key>", "x-rsc-apikey" = "<your API Key>" }

Only include the header(s) for the part(s) of your account you use.

Other clients

Point the client at the endpoint URL using the Streamable HTTP transport, and send x-qrm-apikey and/or x-rsc-apikey as request headers, with your API Key as the value. Check your client's own documentation for where it lets you set custom request headers.

Swap in the Pre Production URL, in any of the above, to test first.

What you can ask your assistant to do

Once connected, ask your assistant what it can do. It will list the current set of actions available to your API Key(s). From there, you can ask it to work with your data in plain language, for example:

• "Look up record <ID> and tell me its current status."
• "Create a new entry with these details: ..."
• "Update <ID> so that <field> is <value>."
• "List everything that changed on <ID> in the last week."
• "Find the open items assigned to <name>."

Your assistant turns each request into the same underlying call you would otherwise make yourself through the API, using the API Key(s) you supplied when you connected.

Operational details

• Timeouts. A call to the underlying service can take up to 120 seconds. A call that takes longer will fail; retry it, or break your request into smaller pieces.
• Errors. If a call fails, the error comes from the underlying service. It is the same error you would see calling that service's API directly, not a separate MCP-specific error.
• Rate limits. The MCP server does not apply its own rate limit. You are subject to whatever limit the underlying integration API already applies to your API Key.
• No separate account. There is no MCP-specific login or account to set up. Access is controlled entirely by the API Key header(s) you send.
• Tool approval. Most MCP clients, including Claude Code and Claude Desktop, ask you to approve a tool the first time your assistant tries to use it, and let you review or revoke that approval later. Check your client's own settings to manage this.

Get started

If you run into an issue, check that:

• Your API Key(s) are active and have not been removed from your account.
• You are using the correct endpoint for your environment (Pre Production vs. Production).
• You are sending the header(s) that match the part(s) of your account you want to reach.
• Your MCP client supports the Streamable HTTP transport and lets you set custom request headers.