########################################################################
# GLOBAL SETTINGS
# These are applied across the entire Helm release.
########################################################################
global:
imagePullSecrets: [] # e.g. ["ghcr-creds"] for a private registry
nameOverride: "" # short name applied to all resources (optional)
fullnameOverride: "" # fully-qualified name override (optional)
########################################################################
# MCP CONTEXT-FORGE (Gateway / API tier)
########################################################################
mcpContextForge:
replicaCount: 2 # horizontal scaling for the gateway
# --- HORIZONTAL POD AUTOSCALER --------------------------------------
# * Percentages compare live usage with the container *request* values
# (limits are ignored by the HPA).
# * If both CPU and memory targets are set, crossing either threshold
# triggers a scale event.
# --------------------------------------------------------------------
hpa:
enabled: true # Set to false to keep a fixed replica count
minReplicas: 2 # Never scale below this
maxReplicas: 10 # Never scale above this
targetCPUUtilizationPercentage: 90 # Scale up when avg CPU > 90 % of *request*
targetMemoryUtilizationPercentage: 90 # Scale up when avg memory > 90 % of *request*
image:
repository: ghcr.io/ibm/mcp-context-forge
tag: latest # pin a specific immutable tag in production
#pullPolicy: IfNotPresent
pullPolicy: Always # always pull the latest image; useful for dev/testing
# Service that fronts the gateway
service:
type: ClusterIP
port: 80 # external port → containerPort below
containerPort: 4444 # port the app listens on inside the pod
# Health & readiness probes
probes:
startup:
# Uncomment to enable sleep startup probe; useful for long-running initializations
type: exec
command: ["sh", "-c", "sleep 10"]
timeoutSeconds: 15 # must exceed the 10-second sleep
periodSeconds: 5
failureThreshold: 1
readiness:
type: http
path: /ready
port: 4444
initialDelaySeconds: 15 # wait 15 s after container start
periodSeconds: 10 # check every 10 s
timeoutSeconds: 2 # fail if no response in 2 s
successThreshold: 1 # one success flips it back to healthy
failureThreshold: 3 # three failures mark pod un-ready
liveness:
type: http
path: /health
port: 4444
initialDelaySeconds: 10 # wait 10 s after container start
periodSeconds: 15
timeoutSeconds: 2
successThreshold: 1
failureThreshold: 3
# Kubernetes resource requests / limits
resources:
limits:
cpu: 200m
memory: 1024Mi
requests:
cpu: 100m
memory: 512Mi
# Optional ingress for HTTP traffic
ingress:
enabled: true
className: nginx
host: gateway.local # CHANGE to your FQDN (e.g. api.example.com)
path: /
pathType: Prefix
annotations:
nginx.ingress.kubernetes.io/rewrite-target: /
####################################################################
# CORE ENVIRONMENT - injected one-by-one as name/value pairs.
# Only the DATABASE / CACHE connection points live here; everything
# else goes into the ConfigMap or Secret blocks below.
####################################################################
env:
host: 0.0.0.0 # bind address inside the container
postgres:
# host is auto-generated as <release>-mcp-stack-postgres
# host: postgres # uncomment to override the generated name
port: 5432
db: postgresdb
userKey: POSTGRES_USER # key in the secret that stores the username
passwordKey: POSTGRES_PASSWORD
redis:
# host is auto-generated as <release>-mcp-stack-redis
# host: redis # uncomment to override the generated name
port: 6379
####################################################################
# PLAIN-TEXT (NON-SECRET) SETTINGS
# Rendered into a ConfigMap; readable by anyone with GET access.
####################################################################
config:
# - Gunicorn settings ─
GUNICORN_WORKERS: "2" # number of worker processes
GUNICORN_TIMEOUT: "600" # worker timeout in seconds
GUNICORN_MAX_REQUESTS: "10000" # max requests per worker before restart
GUNICORN_MAX_REQUESTS_JITTER: "100" # random jitter to avoid thundering herd
GUNICORN_PRELOAD_APP: "true" # preload app code before forking workers (TODO: not implemented yet)
# ─ Basic application info ─
APP_NAME: MCP_Gateway # public-facing name of the gateway
HOST: 0.0.0.0 # address the server binds to
PORT: "4444" # internal container port
APP_ROOT_PATH: "" # e.g. "/gateway" when deploying under sub-path
# ─ Connection pooling ─
DB_POOL_SIZE: "200" # size of SQLAlchemy connection pool
DB_MAX_OVERFLOW: "10" # extra connections allowed beyond pool size
DB_POOL_TIMEOUT: "30" # seconds to wait for a connection
DB_POOL_RECYCLE: "3600" # recycle connections after N seconds
# ─ Cache behaviour ─
CACHE_TYPE: redis # Backend cache driver (redis, memory, database)
CACHE_PREFIX: mcpgw # Prefix applied to every cache key
SESSION_TTL: "3600" # TTL (s) for user sessions
MESSAGE_TTL: "600" # TTL (s) for ephemeral messages (completions)
# ─ Connection retry settings ─
REDIS_MAX_RETRIES: "3" # Maximum retries for Redis cold start
REDIS_RETRY_INTERVAL_MS: "2000" # Interval between Redis retries (ms)
DB_MAX_RETRIES: "3" # Maximum retries for DB cold start
DB_RETRY_INTERVAL_MS: "2000" # Interval between DB retries (ms)
# ─ Protocol & feature toggles ─
PROTOCOL_VERSION: 2025-03-26
MCPGATEWAY_UI_ENABLED: "true" # toggle Admin UI
MCPGATEWAY_ADMIN_API_ENABLED: "true" # toggle Admin API endpoints
CORS_ENABLED: "true" # enable CORS processing in gateway
ALLOWED_ORIGINS: '["http://localhost","http://localhost:4444"]' # JSON list of allowed origins
SKIP_SSL_VERIFY: "false" # skip TLS certificate verification on upstream calls
# ─ Logging ─
LOG_LEVEL: INFO # DEBUG, INFO, WARNING, ERROR, CRITICAL
LOG_FORMAT: json # json or text format
# ─ Transports ─
TRANSPORT_TYPE: all # comma-separated list: http, ws, sse, stdio, all
WEBSOCKET_PING_INTERVAL: "30" # seconds between WS pings
SSE_RETRY_TIMEOUT: "5000" # milliseconds before SSE client retries
# ─ Streaming sessions ─
USE_STATEFUL_SESSIONS: "false" # true = use event store; false = stateless
JSON_RESPONSE_ENABLED: "true" # default to JSON; false for SSE stream
# ─ Federation ─
FEDERATION_ENABLED: "true" # enable federated mode
FEDERATION_DISCOVERY: "false" # advertise & discover peers automatically
FEDERATION_PEERS: '[]' # explicit peer list (JSON array)
FEDERATION_TIMEOUT: "30" # seconds before peer request timeout
FEDERATION_SYNC_INTERVAL: "300" # seconds between peer syncs
# ─ Resource cache ─
RESOURCE_CACHE_SIZE: "1000" # max resources kept in memory cache
RESOURCE_CACHE_TTL: "3600" # TTL (s) for resources in cache
MAX_RESOURCE_SIZE: "10485760" # max allowed resource size in bytes (10 MB)
# ─ Tool limits ─
TOOL_TIMEOUT: "60" # seconds per tool execution
MAX_TOOL_RETRIES: "3" # retries for failed tool runs
TOOL_RATE_LIMIT: "100" # invocations per minute cap
TOOL_CONCURRENT_LIMIT: "10" # concurrent tool executions
# ─ Prompt cache ─
PROMPT_CACHE_SIZE: "100" # number of prompt templates to cache
MAX_PROMPT_SIZE: "102400" # max template size in bytes
PROMPT_RENDER_TIMEOUT: "10" # seconds to render a template
# ─ Health checks ─
HEALTH_CHECK_INTERVAL: "60" # seconds between peer health checks
HEALTH_CHECK_TIMEOUT: "10" # request timeout per health check
UNHEALTHY_THRESHOLD: "3" # failed checks before peer marked unhealthy
FILELOCK_NAME: gateway_healthcheck_init.lock # lock file used at start-up
# ─ Development toggles ─
DEV_MODE: "false" # enable dev-mode features
RELOAD: "false" # auto-reload code on changes
DEBUG: "false" # verbose debug traces
####################################################################
# SENSITIVE SETTINGS
# Rendered into an Opaque Secret. NO $(VAR) expansion here.
# DATABASE_URL & REDIS_URL are declared inside the Deployment
# so their placeholders resolve at runtime. Override them if needed.
####################################################################
secret:
# ─ Admin & auth ─
BASIC_AUTH_USER: admin # username for basic-auth login
BASIC_AUTH_PASSWORD: changeme # password for basic-auth (CHANGE IN PROD!)
AUTH_REQUIRED: "true" # enforce authentication globally (true/false)
JWT_SECRET_KEY: my-test-key # secret key used to sign JWT tokens
JWT_ALGORITHM: HS256 # signing algorithm for JWT tokens
TOKEN_EXPIRY: "10080" # JWT validity (minutes); 10080 = 7 days
AUTH_ENCRYPTION_SECRET: my-test-salt # passphrase to derive AES key for secure storage
# (derived URLs are defined in deployment-mcp.yaml)
# ─ Optional database / redis overrides ─
# DATABASE_URL: "postgresql://admin:s3cr3t@db.acme.com:5432/prod" # override the auto-generated URL
# REDIS_URL: "redis://cache.acme.com:6379/0" # override the auto-generated URL
####################################################################
# Names of ConfigMap / Secret are resolved by templates; leave as-is.
####################################################################
envFrom:
- secretRef:
name: mcp-gateway-secret
- configMapRef:
name: mcp-gateway-config
########################################################################
# DATABASE MIGRATION (Alembic)
# Runs as a Job before mcpgateway deployment
########################################################################
migration:
enabled: true # Set to false to skip migrations
# Job configuration
restartPolicy: Never # Job should not restart on failure
backoffLimit: 3 # Retry up to 3 times before giving up
activeDeadlineSeconds: 600 # Kill job after 10 minutes
# Use same image as mcpgateway
image:
repository: ghcr.io/ibm/mcp-context-forge
tag: latest # Should match mcpContextForge.image.tag
#pullPolicy: IfNotPresent
pullPolicy: Always # always pull the latest image; useful for dev/testing
# Resource limits for the migration job
resources:
limits:
cpu: 200m
memory: 512Mi
requests:
cpu: 100m
memory: 256Mi
# Migration command configuration
command:
waitForDb: "python3 /app/mcpgateway/utils/db_isready.py --max-tries 30 --interval 2 --timeout 5"
migrate: "alembic upgrade head || echo '⚠️ Migration check failed'"
########################################################################
# POSTGRES DATABASE
########################################################################
postgres:
enabled: true
image:
repository: postgres
tag: "17"
pullPolicy: IfNotPresent
service:
type: ClusterIP
port: 5432
# PersistentVolumeClaim for data durability
persistence:
enabled: true
storageClassName: manual # pick a StorageClass (e.g. gp2, standard)
accessModes: [ReadWriteMany]
size: 5Gi
# Leave blank to autogenerate <release>-mcp-stack-postgres-secret.
existingSecret: ""
credentials: # used only when existingSecret is blank
database: postgresdb
user: admin
password: test123 # CHANGE ME in production!
# ─── Resource limits & requests ───
resources:
limits:
cpu: 1000m # 1 core hard cap
memory: 1Gi
requests:
cpu: 500m # guaranteed half-core
memory: 64Mi
# ─── Health & readiness probes ───
probes:
readiness:
type: exec
command: ["pg_isready", "-U", "$(POSTGRES_USER)"]
initialDelaySeconds: 15
periodSeconds: 10
timeoutSeconds: 3
successThreshold: 1
failureThreshold: 3
liveness:
type: exec
command: ["pg_isready", "-U", "$(POSTGRES_USER)"]
initialDelaySeconds: 10
periodSeconds: 15
timeoutSeconds: 3
successThreshold: 1
failureThreshold: 5
########################################################################
# REDIS CACHE
########################################################################
redis:
enabled: true
image:
repository: redis
tag: latest
pullPolicy: IfNotPresent
service:
type: ClusterIP
port: 6379
# ─── Resource limits & requests ───
resources:
limits:
cpu: 100m # cap at 0.1 core, 256 MiB
memory: 256Mi
requests:
cpu: 50m # reserve 0.05 core, 128 MiB
memory: 16Mi
# ─── Health & readiness probes ───
probes:
readiness:
type: exec
command: ["redis-cli", "PING"]
initialDelaySeconds: 10
periodSeconds: 10
timeoutSeconds: 2
successThreshold: 1
failureThreshold: 3
liveness:
type: exec
command: ["redis-cli", "PING"]
initialDelaySeconds: 5
periodSeconds: 15
timeoutSeconds: 2
successThreshold: 1
failureThreshold: 5
########################################################################
# PGADMIN - Web UI for Postgres
########################################################################
pgadmin:
enabled: true
image:
repository: dpage/pgadmin4
tag: latest
pullPolicy: IfNotPresent
service:
type: ClusterIP
port: 80
env:
email: admin@example.com
password: admin123 # CHANGE ME in production!
# ─── Resource limits & requests ───
resources:
limits:
cpu: 200m # cap at 0.2 core, 256 MiB
memory: 256Mi
requests:
cpu: 100m # reserve 0.1 core, 128 MiB
memory: 128Mi
# ─── Health & readiness probes ───
probes:
readiness:
type: http
path: /misc/ping # lightweight endpoint
port: 80
initialDelaySeconds: 15
periodSeconds: 10
timeoutSeconds: 2
successThreshold: 1
failureThreshold: 3
liveness:
type: http
path: /misc/ping
port: 80
initialDelaySeconds: 10
periodSeconds: 15
timeoutSeconds: 2
successThreshold: 1
failureThreshold: 5
########################################################################
# REDIS-COMMANDER - Web UI for Redis
########################################################################
redisCommander:
enabled: true
image:
repository: rediscommander/redis-commander
tag: latest
pullPolicy: IfNotPresent
service:
type: ClusterIP
port: 8081
# ─── Resource limits & requests ───
resources:
limits:
cpu: 100m # cap at 0.1 core, 256 MiB
memory: 256Mi
requests:
cpu: 50m # reserve 0.05 core, 128 MiB
memory: 128Mi
# ─── Health & readiness probes ───
probes:
readiness:
type: http
path: / # root returns 200 OK
port: 8081
initialDelaySeconds: 15
periodSeconds: 10
timeoutSeconds: 2
successThreshold: 1
failureThreshold: 3
liveness:
type: http
path: /
port: 8081
initialDelaySeconds: 10
periodSeconds: 15
timeoutSeconds: 2
successThreshold: 1
failureThreshold: 5
########################################################################
# MCP-FAST-TIME-SERVER - optional high-performance time server for MCP (go)
# Provides a fast implementation including SSE and Streamable HTTP
########################################################################
mcpFastTimeServer:
enabled: true # switch to true to deploy
replicaCount: 2
image:
repository: ghcr.io/ibm/fast-time-server
tag: "0.3.0"
pullPolicy: IfNotPresent
port: 8080
# Ingress example (leave as-is if you already have it)
ingress:
enabled: true
path: /fast-time
pathType: Prefix
servicePort: 80
# ─── Health & readiness probes ───
probes:
readiness:
type: http
path: /health
port: 8080
initialDelaySeconds: 3
periodSeconds: 10
timeoutSeconds: 2
successThreshold: 1
failureThreshold: 3
liveness:
type: http
path: /health
port: 8080
initialDelaySeconds: 3
periodSeconds: 15
timeoutSeconds: 2
successThreshold: 1
failureThreshold: 3
# Tiny Go process: ~10 MB runtime footprint
resources:
limits:
cpu: 50m # ~5 % of a core
memory: 64Mi
requests:
cpu: 25m
memory: 10Mi
