qdrant values¶
Every value of chart qdrant 0.3.1, read from values.yaml and values.schema.json. A map is listed with its keys; a map without a description of its own is listed by its keys only. Every value has a default: the chart installs with none given.
| Value | Type | Default | Description |
|---|---|---|---|
image.repository |
string | docker.io/qdrant/qdrant |
|
image.tag |
string | v1.19.2 |
|
image.pullPolicy |
string | IfNotPresent |
One of: Always, IfNotPresent, Never. |
persistence.size |
string | 5Gi |
|
persistence.storageClass |
string | "" |
Empty: the cluster default StorageClass |
resources.requests.cpu |
string | 100m |
|
resources.requests.memory |
string | 256Mi |
|
resources.limits.cpu |
string | 2 |
|
resources.limits.memory |
string | 2Gi |
|
secrets.type |
string | kubernetes |
kubernetes: a Secret from the value below. externalSecret: from a secret store One of: kubernetes, externalSecret. |
secrets.apiKey |
string | "" |
API key every client sends (api-key header). Empty: generated once and kept on upgrade; a tool that renders without cluster access must set it |
secrets.externalSecret.storeKind |
string | ClusterSecretStore |
One of: ClusterSecretStore, SecretStore. |
secrets.externalSecret.storeName |
string | "" |
|
secrets.externalSecret.refreshInterval |
string | 1h |
|
secrets.externalSecret.items |
object | vault item holding the key in its password field; empty: <release>-qdrant-api-key | |
secrets.externalSecret.items.apiKey |
string | "" |
|
connection |
object | ConfigMap <release>-qdrant-connection with QDRANT_URL and QDRANT_GRPC_URL for clients | |
connection.configMap |
boolean | true |
|
networkPolicy.enabled |
boolean | true |
|
networkPolicy.ingressNamespaces |
array | [] |
Namespaces (label selectors) that may call Qdrant besides its own |
mcp |
object | Memory tools for AI agents: the official mcp-server-qdrant (qdrant-store, qdrant-find) on streamable HTTP, discovered by an MCP gateway. Renders only with mcp.image set: upstream publishes no image, build it from https://github.com/qdrant/mcp-server-qdrant (Dockerfile in the repository) | |
mcp.enabled |
boolean | true |
|
mcp.image.repository |
string | "" |
|
mcp.image.tag |
string | "" |
|
mcp.image.pullPolicy |
string | IfNotPresent |
One of: Always, IfNotPresent, Never. |
mcp.collection |
string | memory |
Collection that holds the memories; one per release |
mcp.embeddingModel |
string | sentence-transformers/all-MiniLM-L6-v2 |
FastEmbed model that turns memories into vectors; fixed once the collection exists |
mcp.modelCache |
object | Volume the model is downloaded to once, instead of at every start | |
mcp.modelCache.enabled |
boolean | true |
|
mcp.modelCache.size |
string | 1Gi |
|
mcp.modelCache.storageClass |
string | "" |
Empty: the cluster default StorageClass |
mcp.port |
integer | 8000 |
Not 80, 443, 8080 or 8443 (open to every project) |
mcp.path |
string | /mcp/ |
Path the gateway calls; mcp-server-qdrant answers /mcp/ (it redirects /mcp) |
mcp.appProtocol |
string | agentgateway.dev/mcp |
Port appProtocol an MCP gateway discovers; empty leaves it unset |
mcp.targetName |
string | memory |
Tool prefix on the gateway (<targetName>_<tool>) |
mcp.findDescription |
string | Long-term team memory. Call this at the start of every task with the task top... |
Tool descriptions: the only guidance the model gets, so they say when to call |
mcp.storeDescription |
string | Save a durable fact to long-term team memory: a decision and its reason, a us... |
|
mcp.extraEnv |
object | {} |
Further mcp-server-qdrant settings as environment variables |
mcp.podLabels |
object | {} |
Extra pod labels, e.g. sidecar.istio.io/inject: "false" in a mesh namespace |
mcp.resources.requests.cpu |
string | 50m |
|
mcp.resources.requests.memory |
string | 384Mi |
|
mcp.resources.limits.cpu |
string | 1 |
|
mcp.resources.limits.memory |
string | 1Gi |
|
mcp.ingressNamespaces |
array | [] |
Namespaces (label selectors) that may call the tools besides the own one: the gateway's |
mcp.allowInternet |
boolean | true |
Let the server reach the internet to download its embedding model (Hugging Face); false for an image with the model built in |