IBM Concert Server

License: Apache 2.0

A Model Context Protocol (MCP) server that connects AI agents and assistants to IBM Concert — enabling autonomous workflow health monitoring, incident diagnosis, auto-remediation, CVE triage, and Slack alerting through natural language.

Overview

This repository contains two complementary MCP servers:

Server	Module	Purpose
concert-workflow-monitor	src/ibm_concert_mcp/server.py	Live workflow health checks, auto-remediation, CVE queries, Slack alerts, incident tracking
concert-knowledge	src/ibm_concert_mcp/knowledge/server.py	Semantic RAG search over IBM Concert documentation (Supabase + pgvector)

Available Tools

concert-workflow-monitor

Tool	Description
list_monitored_workflows	List all Concert workflows being monitored
check_workflow_health	Run a live health check against one or all workflows
get_failing_workflows	Return only workflows currently in ERROR/WARNING state with diagnosis
get_workflow_diagnosis	Diagnose a workflow and auto-post an error alert to Slack
remediate_workflow	Auto-fix a failing workflow (soft re-invoke or oc rollout restart)
get_concert_cve_summary	Query Concert for top CVEs across your application inventory
trigger_package_remediation	Invoke Concert's built-in package remediation workflow (raises GitHub PR + ServiceNow ticket)
upload_scan_to_concert	Push an updated vulnerability/SBOM scan to Concert
send_slack_alert	Post a notification to the configured Slack channel
get_incident_stats	MTTR and incident statistics for the last N days
get_recent_incident_log	Most recent incidents from the incident database
receive_webhook_event	Process inbound Concert webhook events for zero-latency detection
record_recommendation_feedback	Record 👍/👎 engineer feedback to improve future diagnoses
get_recommendation_learnings	Show what the agent has learned from past feedback

concert-knowledge

Tool	Description
search_concert_docs	Semantic similarity search over all ingested Concert documentation
list_concert_sources	List all documentation sources indexed in the knowledge base

Prerequisites

• Python 3.10 or higher
• pip or uv
• Access to an IBM Concert instance
• Slack Incoming Webhook URL (for alerting)
• (Optional) Supabase project for incident tracking and knowledge RAG
• (Optional) OpenShift CLI (oc) for the oc rollout restart remediation strategy

Installation

git clone https://github.com/ibm-ai-platform/ibm-concert-mcp.git
cd ibm-concert-mcp

# Workflow monitor
pip install -r requirements.txt

# Knowledge server (separate — heavier ML dependencies)
pip install -r requirements-knowledge.txt

Or with uv:

uv venv && source .venv/bin/activate
uv pip install -r requirements.txt

Configuration

Copy .env.example to .env and fill in your values:

cp .env.example .env

Required environment variables

Variable	Description
CONCERT_API_KEY	IBM Concert API key (base64-encoded user:token)
SLACK_WEBHOOK_URL	Slack Incoming Webhook URL

Optional environment variables

Variable	Default	Description
CONCERT_BASE_URL	TechZone gateway	Concert Workflows gateway URL
CONCERT_REST_BASE_URL	TechZone host	Concert main host (for CVE/ingestion API)
CONCERT_INSTANCE_ID	—	Concert instance UUID (for CVE API)
SUPABASE_URL	—	Supabase project URL (incident DB + knowledge RAG)
SUPABASE_PUBLISHABLE_KEY	—	Supabase anon/publishable key
OC_NAMESPACE	concert-workflows	OpenShift namespace for oc remediation
RUNBOOK_DIR	./runbooks	Directory where auto-generated runbooks are written
RECOVERY_TIMEOUT_SECONDS	90	How long to wait for recovery after oc restart
ALERT_ON_WARNING	false	Also alert on HTTP 200 empty-body (warning) responses
AUTO_SLACK	true	Auto-post Slack alerts on diagnosis/resolution
VERIFY_TLS	false	Verify TLS certificates
MCP_TRANSPORT	stdio	stdio | sse | streamable-http
MCP_HOST	127.0.0.1	Host for SSE/HTTP transport
MCP_PORT	8000	Port for SSE/HTTP transport (workflow monitor)

For the knowledge server, also set:

Variable	Description
SUPABASE_URL	Supabase project URL
SUPABASE_PUBLISHABLE_KEY	Supabase anon/publishable key

Usage

stdio transport (AI agent / IDE use)

Add the following to your MCP client configuration (e.g. ~/.bob/settings/mcp_settings.json, Claude Desktop claude_desktop_config.json, or VS Code mcp.json):

{
  "mcpServers": {
    "concert-workflow-monitor": {
      "command": "uvx",
      "args": [
        "concert-workflow-monitor"
      ],
      "env": {
        "CONCERT_API_KEY": "<your-concert-api-key>",
        "SLACK_WEBHOOK_URL": "<your-slack-webhook-url>",
        "MCP_TRANSPORT": "stdio"
      }
    },
    "concert-knowledge": {
      "command": "uvx",
      "args": [
        "--from",
        "ibm-concert-mcp[knowledge]",
        "python",
        "-m",
        "ibm_concert_mcp.knowledge.server"
      ],
      "env": {
        "SUPABASE_URL": "<your-supabase-url>",
        "SUPABASE_PUBLISHABLE_KEY": "<your-supabase-key>",
        "MCP_TRANSPORT": "stdio"
      }
    }
  }
}

SSE / HTTP transport (shared/remote deployment)

Run the server centrally so colleagues can point their MCP clients at a URL — no local Python install required:

export CONCERT_API_KEY=<your-key>
export SLACK_WEBHOOK_URL=<your-webhook>
export MCP_TRANSPORT=sse
export MCP_HOST=0.0.0.0
export MCP_PORT=8000
python -m ibm_concert_mcp.server

Colleagues add this to their MCP client config:

{
  "mcpServers": {
    "concert-workflow-monitor": {
      "type": "sse",
      "url": "http://<your-server>:8000/sse"
    }
  }
}

A /health HTTP endpoint is also available at http://<host>:<port>/health for dashboard integration.

Typical incident-response loop

Once connected to an AI agent (e.g. IBM Bob, Claude), you can drive the full loop with natural language:

"Check if any Concert workflows are failing"
→ get_failing_workflows()

"Diagnose the akshay-test-fail workflow"
→ get_workflow_diagnosis("akshay-test-fail")

"Try to fix it automatically"
→ remediate_workflow("akshay-test-fail", strategy="auto")

"Post the result to Slack"
→ send_slack_alert(...)

CVE remediation loop

"Show me the top Priority 1 CVEs"
→ get_concert_cve_summary(priority="1")

"Trigger package remediation for my-app, assign to jsmith"
→ trigger_package_remediation(application_name="my-app", git_repo="...", github_assignee="jsmith")

"Upload the updated SBOM"
→ upload_scan_to_concert(scan_type="package_sbom", ...)

Learning loop

The server accumulates 👍/👎 feedback from engineers to improve future diagnoses:

"The recommendation didn't help — what actually fixed it was restarting the gateway pod"
→ record_recommendation_feedback("akshay-test-fail", vote="down", actual_fix="restart gateway pod")

"What has the agent learned so far?"
→ get_recommendation_learnings()

Registering Concert workflows to monitor

Edit the WORKFLOWS list near the top of src/ibm_concert_mcp/server.py:

WORKFLOWS = [
    # (adapter_name, environment, workflow_name, display_name)
    ("my-adapter", "prod", "my-workflow", "My Workflow Display Name"),
]

Architecture

AI Agent (Bob / Claude / etc.)
        │
        │  MCP (stdio or SSE)
        ▼
concert-workflow-monitor ──► IBM Concert Workflows API
        │                ──► OpenShift (oc rollout restart)
        │                ──► Slack (Incoming Webhook)
        │                ──► Supabase (incident DB)
        │
concert-knowledge ──────────► Supabase pgvector (Concert docs RAG)

License

Apache License 2.0 — see LICENSE.

Contributing

Pull requests welcome. See CONTRIBUTING.md for guidelines.