MCP-OpenStack-Ops
Server Configuration
Describes the environment variables required to run the server.
| Name | Required | Description | Default |
|---|---|---|---|
| OS_CACERT | No | SSL CA certificate path for HTTPS (optional, required for secure HTTPS if using https) | |
| OS_PASSWORD | Yes | OpenStack password (required) | |
| OS_USERNAME | Yes | OpenStack username (required) | |
| FASTMCP_HOST | No | HTTP host address for streamable-http mode | 127.0.0.1 |
| FASTMCP_PORT | No | HTTP port for streamable-http mode | 8080 |
| FASTMCP_TYPE | No | Transport type (stdio, streamable-http) | stdio |
| OS_AUTH_HOST | Yes | OpenStack Identity service host address (required) | |
| OS_AUTH_PORT | Yes | OpenStack Identity service port (required) | |
| OS_INTERFACE | No | Interface type (public, internal, admin) | public |
| MCP_LOG_LEVEL | No | Logging level (DEBUG, INFO, WARNING, ERROR, CRITICAL) | INFO |
| OS_IMAGE_PORT | No | Image service port | 9292 |
| OS_REGION_NAME | No | OpenStack region name | RegionOne |
| OS_VOLUME_PORT | No | Volume service port | 8776 |
| OS_COMPUTE_PORT | No | Compute service port | 8774 |
| OS_NETWORK_PORT | No | Network service port | 9696 |
| OS_PROJECT_NAME | Yes | OpenStack project name (required) | |
| OS_AUTH_PROTOCOL | No | Connection protocol (http or https) | http |
| OS_PLACEMENT_PORT | No | Placement service port | 8780 |
| REMOTE_SECRET_KEY | No | Secret key for Bearer token authentication (required if REMOTE_AUTH_ENABLE=true) | |
| OS_HEAT_STACK_PORT | No | Heat orchestration service port | 8004 |
| REMOTE_AUTH_ENABLE | No | Enable Bearer token authentication for streamable-http mode | false |
| OS_USER_DOMAIN_NAME | No | User domain name | default |
| OS_HEAT_STACK_CFN_PORT | No | Heat CloudFormation service port | 18888 |
| OS_PROJECT_DOMAIN_NAME | No | Project domain name | default |
| ALLOW_MODIFY_OPERATIONS | No | Enable modify operations (true/false) | false |
| OS_IDENTITY_API_VERSION | No | Identity API version | 3 |
Instructions
Guidance the server publishes about itself, which clients place ahead of the tool catalog so the model reads it before choosing anything.
MCP-OpenStack-Ops operates one OpenStack project through real OpenStack API calls: compute, networking, block storage, images, identity, Heat, Octavia load balancers, Barbican secrets, quotas, Gnocchi telemetry, and Aodh alarms.
Scope and availability
Every tool call is a real API request scoped to the configured project (OS_PROJECT_NAME). Use the tools for facts; never guess, simulate, or invent results.
set_* tools exist only when the server sets ALLOW_MODIFY_OPERATIONS=true. If a needed set_* tool is missing, say that modifications are disabled.
Some read tools accept all_projects=true, which works only when the server sets ALLOW_CROSS_PROJECT_READ=true. Use it only when the user asks about all or other projects, group results by project_name, and mention scope.warning when present. Report a disabled-switch or policy-denial error as-is; never present current-project results as cloud-wide. Modifications and lookups by name stay in the current project, except get_project_details with all_projects=true, which accepts another project's name or ID.
Optional services (Octavia, Barbican, Heat, Gnocchi, Aodh, port forwarding) may be absent from the cloud. Report the returned error; do not claim the feature works.
Reporting results
Report what the tool returned. Never claim success after an error, an empty response, a partial result, or reported failures; quote the error.
Many operations are asynchronous: an accepted request is not a completed one. Say so, and confirm the final state with a read tool when it matters.
Changing resources
Confirm destructive actions (delete, stop, rebuild, detach) with the user before running them.
When the user targets resources by pattern ("all test instances") for a destructive action, list the matches with a read tool first. set_* tools accept comma-separated names, a JSON array, or filters such as name_contains.
If required creation parameters are missing (for example flavor, image or boot volume, network, size, CIDR), ask for them instead of attempting the operation.
To open a port, ask for the source CIDR or remote group; never assume 0.0.0.0/0 or ::/0.
Never ask for passwords in chat. set_users reads a password from the server environment variable named by password_env.
Never ask for or accept private keys, certificates, or other secret payloads in chat; no tool takes or returns them. Users register their own certificates in Barbican outside MCP; refer to them by name or href.
Numbers and monitoring
Allocation, quota, and capacity (get_resource_monitoring, get_quota, get_hypervisor_details, get_usage_statistics) are not measured utilization. Measured values come only from Gnocchi (get_gnocchi_measures, get_gnocchi_aggregates).
Label every number as allocated, quota, capacity, or measured. Never divide project allocations by physical capacity. Missing capacity is null, not zero. If a metric is not collected, say "not collected"; never estimate it.
Gnocchi metrics vcpus, memory, and disk.root.size record flavor sizes, not usage.
Large projects
Instance and load balancer listings paginate (50 per request by default, at most the server's LIST_MAX_LIMIT). Use limit and offset; use include_all only when the user needs every item.
Capabilities
Features and capabilities supported by this server
Protocol revision2025-11-25
| Capability | Details |
|---|---|
| tools | {
"listChanged": true
} |
| logging | {} |
| prompts | {
"listChanged": true
} |
| resources | {
"subscribe": false,
"listChanged": true
} |
Tools
Functions exposed to the LLM to take actions
| Name | Description |
|---|---|
| get_availability_zonesB | List availability zones and their status |
| get_domainsB | List Keystone domains subject to cloud policy. |
| get_floating_ip_poolsA | Get list of floating IP pools (external networks). Functions:
Use when user requests:
|
| get_floating_ipsB | Get list of floating IPs with their associations. Functions:
Use when user requests floating IP information, external connectivity queries, or IP management tasks. |
| get_gnocchi_aggregatesA | Aggregate measured telemetry across many current-project resources. Use for cross-resource questions: project-wide averages or totals, top N resources, per-flavor or per-host groups, and ratios between metrics. For one resource's time series, use get_gnocchi_measures instead. For allocations, capacity, or quotas, use get_resource_monitoring. Choose ONE mode:
Units: cpu with rate:mean is CPU nanoseconds per granularity window; utilization fraction = value / (granularity * 1e9 * vCPUs). memory.usage is MB in use. vcpus and memory are flavor sizes, not usage. Results are always limited to resources of the current project. The response echoes the executed operations expression. |
| get_gnocchi_measuresA | Read stored Gnocchi measures (telemetry time series) for a metric. Use for measured utilization, load, trends, and history ("CPU usage of vm1", "memory over the last hour"). Values are observed telemetry collected by Ceilometer, such as cpu, memory.usage, or disk and network counters. Cumulative counters like cpu (nanoseconds) grow over time; compare consecutive values for rates. Some metrics such as vcpus and memory record flavor sizes, not usage. For averages, totals, top N, or groups across many resources, use get_gnocchi_aggregates. For allocations, capacity, or quotas, use get_resource_monitoring or get_quota. If a metric is not collected, report that; do not substitute allocation data. |
| get_gnocchi_metricsA | List or show current-project Gnocchi metrics. Without arguments, lists metric names and IDs of current-project resources. With resource_id, shows that resource's metrics including archive policy and unit. With metric, shows one metric. Use it to check which telemetry metrics are actually collected before reading measures or creating alarms. These are measured telemetry series, not allocations or quotas. |
| get_gnocchi_resourcesA | List, search, or show current-project Gnocchi (telemetry) resources. Resources are the objects Ceilometer measures, such as instances, disks, and network interfaces. Each resource maps metric names to metric IDs. Use this to find which metrics exist for a server before reading measures. |
| get_gnocchi_service_infoA | Read Gnocchi service configuration and health information. Use to check which granularities and aggregation methods a metric supports (for get_gnocchi_measures or set_alarms), which resource types exist, and whether measure processing is backlogged. |
| get_heat_stack_detailsB | Query resources, events, template, or outputs for a Heat stack. |
| get_heat_stacksA | Get list of Heat orchestration stacks. Functions:
Use when user requests stack information, orchestration queries, or infrastructure-as-code status. |
| get_hypervisor_detailsC | Get detailed information about hypervisors |
| get_identity_groupsC | List Keystone groups subject to cloud policy. |
| get_image_detail_listB | Get detailed list of all images with comprehensive metadata. Functions:
Use when user requests image listing, image information, or image metadata details. |
| get_image_membersC | List member projects for an image. |
| get_instanceA | Unified instance query tool supporting all instance retrieval patterns. Consolidates functionality from get_instance_details, get_instance_by_name, get_instances_by_status, and search_instances. Functions:
Use when user requests instance information, status checks, or instance searches. |
| get_instance_by_nameB | Get detailed information for a specific instance by name. |
| get_instance_detailsB | Provides detailed information and status for OpenStack instances with pagination support. Functions:
Use when user requests specific instance information, VM details, server analysis, or instance troubleshooting. |
| get_instances_by_statusC | Get instances filtered by status. |
| get_key_manager_quotasA | Show the effective Barbican quotas (secrets, orders, containers, consumers, CAs) of the current project; -1 means unlimited. |
| get_keypair_listA | Get list of SSH keypairs for the current user. Functions:
Use when user requests SSH key management, keypair information, or security key queries. |
| get_load_balancer_amphoraeA | Get amphora instances for a load balancer or all amphorae. |
| get_load_balancer_availability_zonesC | Get load balancer availability zones. |
| get_load_balancer_detailsA | Get detailed information about a specific OpenStack load balancer. Functions:
Use when user requests:
|
| get_load_balancer_flavorsC | Get load balancer flavors. |
| get_load_balancer_health_monitorsA | Get health monitors, optionally filtered by pool. Functions:
Use when user requests:
|
| get_load_balancer_l7_policiesA | Get L7 policies for a listener or all L7 policies. |
| get_load_balancer_l7_rulesC | Get L7 rules for a specific L7 policy. |
| get_load_balancer_listA | Retrieve comprehensive list of OpenStack load balancers with detailed information. Functions:
Use when user requests:
|
| get_load_balancer_listenersA | Get listeners for a specific OpenStack load balancer. Functions:
Use when user requests:
|
| get_load_balancer_pool_membersA | Get members for a specific OpenStack load balancer pool. Functions:
Use when user requests:
|
| get_load_balancer_poolsA | Get load balancer pools, optionally filtered by listener. Functions:
Use when user requests:
|
| get_load_balancer_providersC | Get load balancer providers. |
| get_load_balancer_quotasA | Get load balancer quotas for a project or all projects. |
| get_network_detailsA | Provides detailed information for OpenStack networks, subnets, routers, and security groups. Functions:
Use when user requests network information, subnet details, router configuration, or network troubleshooting. |
| get_network_qos_policiesA | List project-owned Neutron QoS policy containers; no bandwidth rules are created. |
| get_project_detailsC | Get OpenStack project details (similar to 'openstack project list/show'). |
| get_quotaA | Get quota information for projects (similar to 'openstack quota show'). Each service reports limits, usage and usage_percent (percent of each limit in use). A usage value is null when it could not be determined, and usage_percent is null for unlimited (-1), zero or missing limits and for unknown usage. |
| get_resource_monitoringA | Read current project allocations and visible hypervisor capacity. Functions:
Use for current inventory and capacity planning. No historical trends or telemetry samples are collected. Physical totals and project allocations have different scopes and should not be divided into a utilization rate. For measured CPU/memory utilization, load, or trends, use get_gnocchi_measures instead; never present these allocations as usage. |
| get_role_assignmentsA | Get role assignments for the current project. Functions:
Use when user requests permission information, access control queries, or security auditing. |
| get_routersB | Get list of routers with their configuration. Functions:
Use when user requests router information, network connectivity queries, or routing configuration. |
| get_secret_aclA | Show the read ACL of a Barbican secret or container. The ACL lists users outside the owning project allowed to read it and whether project-wide access (project-access) is enabled. |
| get_secret_containersB | List or show Barbican secret containers of the current project. A container groups secret hrefs, e.g. a certificate container with certificate, private_key and intermediates entries. Payloads are never returned. |
| get_secret_ordersA | List or show Barbican key-generation orders of the current project. An ACTIVE order points to the generated secret (key orders) or container (asymmetric orders) through secret_ref or container_ref. |
| get_secret_storesA | List Barbican secret store backends. Only clouds that enable multiple secret store backends serve this API; others return an error. |
| get_secretsA | List or show Barbican secrets of the current project (metadata only). Never returns a secret payload; payload retrieval is intentionally not provided so key material cannot reach the LLM. Use the secret_ref from this tool as a reference, for example in set_load_balancer_listener. |
| get_security_groupsA | Get list of security groups with their rules. Functions:
Use when user requests security group information, firewall rules, or network security queries. |
| get_server_eventsA | Get recent events for a server by name or UUID. Pass either value using instance_name, for example get_server_events(instance_name=""). There is no server_id parameter; do not use server_id or instance_id as argument names. |
| get_server_groupsC | List all server groups with their details |
| get_server_volumesC | Get all volumes attached to a specific server |
| get_service_statusA | Provides status and health check information for each OpenStack service. Functions:
Use when user requests service status, API status, health checks, or service troubleshooting. |
| get_usage_statisticsA | Get current-project Nova allocation-hours and separate current snapshots. Functions:
Use for project usage reports. Period data comes from Nova simple tenant usage. It measures allocation-hours, not live utilization or monetary charges. For measured CPU/memory utilization or trends, use get_gnocchi_measures. Missing or denied usage API returns failure without synthetic fallback. |
| get_user_listA | Get list of OpenStack users in the current domain. Functions:
Use when user requests user management information, identity queries, or user administration tasks. |
| get_volume_backupsB | List Cinder backups visible in the current project. |
| get_volume_groupsA | List current-project Cinder generic groups. Requires API microversion >= 3.13. |
| get_volume_listA | Get list of all volumes with detailed information. Functions:
Use when user requests volume listing, volume information, or storage overview. |
| get_volume_qosA | List Cinder QoS specifications, or show one specification's volume type associations by name or ID. Cloud policy may require admin rights. |
| get_volume_snapshotsA | Get list of volume snapshots. Functions:
Use when user requests snapshot information, backup queries, or volume restoration planning. |
| get_volume_transfersA | List current-project Cinder volume transfer requests without authorization keys. |
| get_volume_typesA | Get list of volume types with their specifications. Functions:
Use when user requests volume type information, storage backend queries, or volume creation planning. |
| search_instancesB | Search for OpenStack instances based on various criteria with efficient pagination. Functions:
|
Prompts
Interactive templates invoked by user choice
| Name | Description |
|---|---|
No prompts | |
Resources
Contextual data attached and managed by the client
| Name | Description |
|---|---|
No resources | |
TDQS
Scored across 61 tools
The instance tools overlap heavily: get_instance is explicitly a unified replacement for get_instance_details, get_instance_by_name, get_instances_by_status, and search_instances, yet all of those legacy tools remain exposed. Other areas are better separated, but the retained redundant instance tools create multiple unclear boundaries for an agent.
Almost all tools use a predictable get_snake_case pattern, with consistent resource grouping such as get_load_balancer_* and get_volume_*. The main deviation is search_instances, and some names awkwardly append list/detail/list, but overall naming is readable and mostly consistent.
With 61 tools, the surface is far beyond the typical well-scoped range and well into the 50+ extreme-mismatch category. Granular read-only queries for load balancer subresources, Gnocchi telemetry, Barbican secrets, and many OpenStack services make the toolset unwieldy for an agent.
The tool surface is almost entirely read-only get/search operations, with no create, update, delete, or lifecycle action tools for instances, networks, volumes, load balancers, Heat stacks, secrets, keypairs, or users. For an OpenStack ops server, this is a severe gap that prevents most operational workflows.