The Model Context Protocol (MCP) is rapidly becoming the standard way to connect AI tools to external data sources and services. IBM has released a preview of the API Connect MCP Server at github.com/ibm-apiconnect/apic-mcp-server, which lets AI assistants and agents query your API Connect analytics data directly — no manual report generation, no custom API wrappers. If you’re building AI-powered workflows that need API programme data, this is a compelling integration point.

In this article, I’ll walk you through what the MCP server is, how to install it, how to connect it to your AI client of choice, and what analytics tools it exposes.

Table of Contents

  1. What Is the MCP Server?
  2. Architecture Overview
  3. Prerequisites
  4. Installation
  5. Configuration
  6. Connecting Your AI Client
  7. Available Analytics Tools
  8. Example Sessions
  9. Debug Logs
  10. Preview Caveats

What Is the MCP Server?

The API Connect MCP Server is a standalone service that implements the Model Context Protocol server interface. MCP, originally developed by Anthropic, defines a standard way for AI applications to connect to external tools and data sources. When an MCP server is running, AI assistants that support MCP can discover its tools and call them as part of a conversation.

The server is distributed as a set of .tgz npm packages (and .mcpb installer files for Claude Desktop), each covering a different capability area. The Analytics service — the focus of this article — exposes your API Connect analytics data as MCP tools. The repository also contains Management and other service packages.

This means you can ask an AI assistant (Claude Desktop, VS Code Copilot, IBM Bob, or any MCP-compatible AI tool) natural language questions about your API programme and get structured answers pulled directly from live API Connect analytics data.

Architecture Overview

┌─────────────────────────┐
│  AI Assistant (Client)  │  Claude Desktop / VS Code / IBM Bob
└────────────┬────────────┘
             │ stdio (MCP protocol)
┌────────────▼────────────┐
│  apic-analytics-        │  ← This is what you install (tgz / mcpb)
│  mcp-server (Node.js)   │
└────────────┬────────────┘
             │ HTTPS REST API
┌────────────▼────────────┐
│  API Connect            │
│  Platform + Management  │
│  Endpoints              │
└─────────────────────────┘

The server is invoked via npx by the MCP client — you do not run it as a persistent background service.

Prerequisites

Before installing, ensure you have:

  1. A supported API Connect deployment with analytics data available to the MCP package you intend to use
  2. Node.js 20+ up to Node.js 24 ("node": ">=20.0.0 <25.0.0" per package.json)
  3. npm installed
  4. The following values from your API Connect instance:
    • Provider Organization name
    • API Key
    • Client ID configured on your instance
    • Client Secret configured on your instance
    • APIC Platform Endpoint URL
    • APIC Management Endpoint URL
  5. A compatible MCP client such as Claude Desktop, VS Code, or IBM Bob

Note: The post originally stated Node.js 18+. The actual package.json in the published package requires Node.js 20–24.

Installation

The preview package is distributed from the GitHub repository as a pre-built .tgz file. Download the current analytics package from:

https://github.com/ibm-apiconnect/apic-mcp-server/tree/main/analytics

The exact filename and version may change between preview releases.

The MCP client invokes it via npx pointing at your local .tgz path — there is no global npm install step.

Configuration

The required configuration is passed entirely via environment variables — there is no separate config file. The environment variables are:

Variable Description
PROVIDER_ORG Your provider organization name
API_KEY Your API Connect API key
client_id Client ID for your APIC instance
client_secret Client Secret for your APIC instance
APIC_PLATFORM_URL Your APIC platform endpoint URL
APIC_MANAGEMENT_URL Your APIC management endpoint URL
NODE_TLS_REJECT_UNAUTHORIZED Set to 0 to disable certificate validation (for self-signed certs), 1 to enable
LOG_LEVEL Optional. Set to debug for verbose logging (default: info)

Security note: Treat API_KEY and client_secret as passwords. Do not commit them to source control. MCP client config files support secret prompting — use that mechanism rather than hardcoding values.

Connecting Your AI Client

The repository includes examples for several MCP clients. In all cases you provide the path to the downloaded .tgz file.

Claude Desktop

If the release includes a .mcpb installer file, Claude Desktop can use it to simplify setup:

  1. Download analytics/apic-analytics-mcp-server-0.0.1.mcpb from the repository
  2. Double-click the .mcpb file — Claude Desktop opens and starts the installation wizard
  3. The wizard prompts for your APIC credentials (see Prerequisites above)
  4. Click Save / Install
  5. Confirm the extension is enabled after installation before testing it

VS Code (GitHub Copilot)

Use the template from the repository at analytics/mcp.vscode.json:

{
  "servers": {
    "apic-analytics-mcp-server": {
      "command": "npx",
      "args": ["-y", "-p", "${input:tarPath}", "apic-analytics-mcp-server"],
      "env": {
        "PROVIDER_ORG": "${input:pOrg}",
        "API_KEY": "${input:api-key}",
        "client_id": "${input:cID}",
        "client_secret": "${input:cSecret}",
        "APIC_PLATFORM_URL": "${input:platformUrl}",
        "APIC_MANAGEMENT_URL": "${input:managementUrl}",
        "NODE_TLS_REJECT_UNAUTHORIZED": "${input:validateCertificates}"
      }
    }
  }
}

Copy this to your workspace .vscode/mcp.json and fill in values, or use VS Code’s inputs prompting to have it ask for credentials interactively. Full instructions: VS Code MCP server setup.

IBM Bob

Use the template from analytics/mcp.bob.json:

{
  "mcpServers": {
    "apic-analytics-mcp-server": {
      "command": "npx",
      "args": ["-y", "-p", "<path-to-tgz>", "apic-analytics-mcp-server"],
      "env": {
        "NODE_TLS_REJECT_UNAUTHORIZED": "1",
        "PROVIDER_ORG": "<your-provider-organization-name>",
        "API_KEY": "<your-api-key>",
        "client_id": "<your-client-id>",
        "client_secret": "<your-client-secret>",
        "APIC_PLATFORM_URL": "<your-apic-platform-url>",
        "APIC_MANAGEMENT_URL": "<your-apic-management-url>"
      }
    }
  }
}

Copy this to .bob/mcp.json in your workspace and replace the placeholder values. Restart Bob after saving. Full instructions: IBM Bob MCP server setup.

Available Analytics Tools

The analytics MCP server exposes analytics-focused tools such as the following (tool names should be checked against the exact preview package you downloaded):

Tool What it answers
GetAnalyticsUsage API call counts and error rates; spikes in API usage; popularity ranking
GetAnalyticsApplication Which applications are driving traffic; application error rates; subscription information
GetAnalyticsConsumer Consumer organization activity; error rates per consumer org; traffic over time
GetAnalyticsLatency API response times (min / max / average); fastest and slowest APIs; time-to-error
GetAnalyticsStatus HTTP response status code distribution across API calls
GetAnalyticsUsers Geographic usage (countries, timezones); user agents; client IP data
GetAnalyticsAILLM AI / LLM token consumption; model usage patterns; consumer orgs using AI services
GetAnalyticsMCP MCP call volumes; MCP tool popularity; consumer orgs using MCP services

Query interpretation and default date handling depend on the MCP client and package version. Check the package documentation or try a few basic queries in your client to confirm the current behaviour.

Some preview builds may also include governance, scan, or OpenAPI-related tools, but those are beyond the scope of this article.

Example Sessions

Example 1: “What were our top APIs by traffic last week?”

What were the top 5 APIs by traffic volume in the last 7 days?

A client could use an analytics usage tool to retrieve and rank API call counts, returning a summary like:

Based on API Connect analytics for the last 7 days:

Top 5 APIs by call volume:
1. /forex/rates       — 2.1M calls
2. /payments/checkout — 1.8M calls
3. /accounts/balance  — 1.2M calls
4. /orders/history    — 890K calls
5. /users/profile     — 654K calls

Example 2: “Are there APIs with elevated error rates?”

Which APIs have error rates above 1% this month?

The client can call one or more analytics tools and return a summary of APIs exceeding the threshold.

Example 3: “Which consumer orgs are sending the most traffic?”

Show me the top consumer organizations by API calls this quarter.

The client can call a consumer-oriented analytics tool and return a ranked list with call counts per consumer org.

Debug Logs

When running, the preview package may write INFO-level logs to a rotating daily file such as:

  • macOS / Linux: ~/apic-mcp/apic-mcp-YYYY-MM-DD.log
  • Windows: %USERPROFILE%\apic-mcp\apic-mcp-YYYY-MM-DD.log

Retention and log-level behaviour may vary by package version. If supported by your build, add LOG_LEVEL: 'debug' to the env section of your MCP client config to enable more verbose logging.

Preview Caveats

As this is a preview (v0.0.1) release, be aware of the following:

  • GitHub-distributed preview package: You point npx at the downloaded .tgz file rather than relying on a global install.
  • Node.js version constraint: Check the package.json bundled with the package you downloaded and use that as the source of truth.
  • Authentication model: Confirm the required credentials for your target package and deployment before rollout.
  • No persistent server process: The package is invoked by the MCP client on demand rather than run as a long-lived background service.
  • Preview support posture: Check IBM release notes or repository guidance for the current support expectations of the preview.
  • MCP spec evolution: As MCP evolves, config formats and tool signatures may change between releases. Check the GitHub repository for the latest version.

Summary

The IBM API Connect MCP Server preview is a promising way to query analytics data from MCP-capable clients. The analytics tools map well to real-world questions about API usage, consumer behaviour, latency, and error rates. If you’re building AI-powered operations workflows, it’s worth evaluating with the exact preview package and client combination you intend to support.

The key practical points: download the package from GitHub, check the bundled package.json for the supported Node.js range, and configure it via environment variables or the client template provided with the release.

See also: IBM API Connect Analytics tools documentation