IBM/mcp — Summary

IBM MCP is IBM's official catalog repository for Model Context Protocol server implementations — a curated list of 30+ MCP servers across IBM's entire product portfolio, covering enterprise automation (IBM MQ, Business Automation Workflow), data analytics (watsonx.data, OpenPages), security (Guardium, QRadar, Security Verify), developer productivity (Carbon Design System, z/OS Developer), and infrastructure (IBM Cloud, Terraform IBM Modules).

Problem it solves: IBM enterprise customers using AI agents lack standardized integrations into IBM's broad product ecosystem. This catalog provides the discovery layer and install instructions for connecting any MCP-compatible AI agent to IBM products — from mainframe (z/OS) to cloud (IBM Cloud) to data platforms (watsonx.data) to security (Guardium).

Distinctive trait: The only vendor MCP catalog in the corpus covering mainframe (z/OS, IBM i) integration through MCP — IBM extends AI agent reach to COBOL/PL/I/JCL environments that no other vendor in this batch addresses. The mcp.json manifest file and minimal repo structure make this the leanest vendor catalog (no source code, just a manifest + README catalog).

Target audience: IBM enterprise customers deploying AI agents who need to connect to IBM products; developers building IBM-cloud and IBM-platform integrations.

Scope: 379 GitHub stars, Apache-2.0, 85 forks, primarily a catalog (no source code beyond mcp.json + README), active since April 2025.

Differs from seeds: Unlike ms-mcp-servers (which hosts source code for Azure/Fabric MCP servers), IBM/mcp is exclusively a catalog pointing to external IBM repos. Unlike any seed framework, IBM/mcp is an enterprise ecosystem navigator — no methodology, no skills, no workflow — just a curated index with VS Code install badges.

IBM/mcp — Overview

Origin

Published by IBM GitHub organization in April 2025. IBM's homepage for the MCP catalog: https://www.ibm.com/think/topics/model-context-protocol.

Philosophy

"A collection of Model Context Protocol (MCP) servers, clients and developer tools by IBM. Connect your IBM products to any AI Agent or AI application."

Key principles:

1. Universal connectivity: "Connect your IBM products to any AI Agent" — the goal is breaking down AI integration silos around IBM's product portfolio
2. Standard protocol: MCP as the universal integration layer for IBM enterprise products
3. VS Code ecosystem focus: Most servers have VS Code one-click install badges

IBM Product Domains Covered

Third-Party Inclusion

IBM's catalog includes non-IBM servers where they integrate well with IBM products:

• DataStax Astra DB
• HashiCorp Consul + Vault + Terraform
• Langflow

Mainframe Presence

IBM uniquely covers mainframe integration:

• IBM Developer for z/OS: COBOL, PL/I, REXX, JCL, Assembler through VS Code
• IBM i MCP Server: SQL-based tools for IBM i systems
• z/OS Connect: IBM APIs exposed through MCP

IBM/mcp — Architecture

Distribution

• Type: methodology-doc (catalog only — no source code)
• License: Apache-2.0
• Structure: README catalog + mcp.json manifest

Repository Structure

IBM/mcp/
├── mcp.json              # MCP catalog manifest
├── README.md             # Catalog with VS Code install badges
└── CONTRIBUTING.md       # How to add IBM MCP servers to catalog

Catalog Only — No Source Code

Unlike microsoft/mcp (which hosts Azure/Fabric server source in C#), IBM/mcp contains:

• Zero lines of implementation code
• mcp.json — a manifest file listing catalog entries
• README — the human-readable catalog with install instructions
• CONTRIBUTING.md — process for adding new servers

All MCP server implementations are in external IBM GitHub repos linked from the README.

mcp.json Structure

The mcp.json file is the machine-readable catalog manifest. Format (inferred from README listing structure):

{
  "servers": [
    {
      "name": "ibm-mq-mcp-server",
      "repo": "https://github.com/ibm-messaging/mq-mcp-server",
      "category": "Automation",
      ...
    },
    ...
  ]
}

External Repo Pattern

Each listed server follows a pattern:

1. Separate IBM GitHub repo (e.g., github.com/IBM/gdp-mcp-server)
2. VS Code install badge (one-click install URL)
3. Optional standalone documentation link

Installation Patterns

Different servers offer different install methods:

• VS Code extension: One-click vscode:mcp/install?{...} URLs
• uvx: uvx --from git+https://github.com/... server-name
• npx: npx -y @package/name
• Docker: docker run image:tag
• See link: Some complex servers link to their own documentation

Required Runtime

Varies by server:

• Python + uvx (most Python-based servers)
• Node + npx (TypeScript servers)
• Docker (containerized servers)
• IBM product credentials (IBM Cloud API key, watsonx endpoint, etc.)

IBM/mcp — Components

MCP Servers by Category (30+)

Automation (2)

Business Automation (5)

Data & Analytics (8)

Developer Productivity (3)

Infrastructure & Deployment (7)

Observability & Monitoring (2)

Security (6)

Developer Tools (5)

Research (1)

Notable: ContextForge MCP Gateway

"MCP server, feature-rich gateway, and proxy that federates MCP and REST services — delivering unified discovery, authentication, rate-limiting, observability, multi-transport protocols, and an optional HTMX-powered admin UI through a single endpoint."

This is the most sophisticated server in IBM's catalog — a meta-gateway that unifies multiple MCP and REST services behind a single endpoint with auth, rate limiting, and a web admin UI.

IBM/mcp — Uniqueness & Positioning

Differs from Seeds

IBM/mcp is categorically different from all 11 seeds — it is an enterprise product catalog, not a workflow framework. The closest analogy is how the ccplugins-awesome repo serves as a community plugin catalog, but IBM/mcp is:

• Vendor-curated (IBM only)
• Enterprise-grade products
• MCP servers (not Claude Code plugins)
• Includes mainframe integration

Among vendor MCP catalogs in the batch, IBM/mcp is the leanest (no source code) while microsoft/mcp hosts source. IBM's approach is purely catalog-and-pointer.

Distinctive Properties

1. Mainframe integration: IBM Developer for z/OS MCP + IBM i MCP + z/OS Connect — the only vendor in the batch covering COBOL/PL/I/JCL environments through AI agent tooling. This addresses a unique enterprise segment (IBM mainframe shops) ignored by all other frameworks.

2. Biomedical AI: MAMMAL-MCP (protein-protein interaction, drug-protein interaction) — the only MCP server in the entire batch targeting scientific/biomedical AI workloads.

3. Security depth: 4 security servers (Guardium DP, QRadar SIEM, Guardium Cryptography, Security Verify) covering the full enterprise security stack — more security coverage than any other vendor catalog.

4. 4-tool meta-API pattern: Guardium DP (621 APIs → 4 tools), QRadar (REST API → 4 tools), Security Verify (210 APIs → 4 tools) — consistent pattern of compressing large API surfaces into small tool counts for LLM friendliness.

5. Minimal repo size: The entire catalog is a README + mcp.json — zero source code. This is the leanest architecture in the corpus for maximum catalog scope.

Observable Failure Modes

• Catalog staleness: External repos can change, get deprecated, or go unmaintained while the IBM/mcp catalog still lists them
• Fragmentation: 30+ separate repos, each with different install patterns, credentials, and documentation quality
• IBM product complexity: Many servers require IBM Cloud credentials, endpoints, or license agreements that create onboarding friction
• No integration testing: No way to know if any listed server still works without actually installing it

The ContextForge Opportunity

The ContextForge MCP Gateway (IBM Research contribution) is the most architecturally interesting entry in IBM's catalog — it federates multiple MCP + REST services, provides auth, rate limiting, and an admin UI. It represents what a "unified IBM agent integration layer" could look like, but it is not currently an IBM-managed product.