vbr-write-mcp
Server Configuration
Describes the environment variables required to run the server.
| Name | Required | Description | Default |
|---|---|---|---|
| MCP_TOKENS | No | JSON object mapping long random bearer tokens to team names, used for HTTP transport authentication. | |
| VBRW_DRY_RUN | No | Set to true to avoid sending any write requests and only show the requests that would be sent. | |
| VBR_BASE_URL | Yes | The URL of the Veeam Backup & Replication REST API server, for example http://host:9419. | |
| VBR_PASSWORD | Yes | The VBR password used to authenticate with the Veeam Backup & Replication REST API. | |
| VBR_USERNAME | Yes | The VBR username used to authenticate with the Veeam Backup & Replication REST API. | |
| VBRW_AUDIT_LOG | No | Path for the JSONL audit log. Every attempt is recorded with plan, confirmation type, result, error, and client details, with secrets masked. | |
| VBRW_HTTP_PORT | No | HTTP port used when VBRW_TRANSPORT=http, for example 8001. | |
| VBRW_TRANSPORT | No | Transport mode. Set to http for shared HTTP usage; otherwise the default stdio transport is used. | |
| VBRW_TOOL_GROUPS | No | Narrows the exposed tool groups. | |
| VBRW_ACCESS_LEVEL | No | Access level for the server. Use write to hide destructive tools and reject dangerous variants of mixed tools; recommended for production. | |
| VBRW_BLOCKED_PATHS | No | Additional paths blocked for api_write. The paths /api/oauth2/* and registerVbr are always blocked. | |
| VBRW_DISABLED_TOOLS | No | Comma-separated or otherwise configured list of tool names to disable. | |
| VBRW_CONFIRMATION_MODE | No | Server-side confirmation mode. Valid values are auto (default), elicit, token, and none. The none mode disables confirmation and is not recommended. | auto |
Instructions
Guidance the server publishes about itself, which clients place ahead of the tool catalog so the model reads it before choosing anything.
Write and administration tools for one Veeam Backup & Replication (VBR) server over its REST API (1.3-rev2). They
complement Veeam's read-oriented vbr-mcp: use vbr-mcp (or api_read here) to find ids and check state, and these
tools to change things.
Every state-changing tool goes through a server-side safety layer:
Destructive tools (delete, restore, failover/failback, stopping sessions, edits of existing objects, settings, removing licenses, marking malware clean...) are confirmed by the server. Either the user is asked directly by the MCP client, or the tool returns
status: confirmation_requiredwith a plan and aconfirm_token. In that case show the plan to the user in plain words (target name, what will happen, the impact lines), ask for an explicit yes, and only then call the same tool again with the same arguments plusconfirm_token. Never invent or reuse a token, never treat the user's original request as the yes, and never confirm on their behalf.status: dry_runmeans the server is in dry-run mode and nothing was sent.status: submittedmeans VBR accepted the request. Most actions are asynchronous: say they were started, then follow the returned session withget_session_result.
Creating or configuring something for the user (a backup job, repository, credentials, ...): never ask the user
for JSON or field names. Ask in plain language for what is missing, a few questions at a time, and offer concrete
choices read from the server instead of open questions (the VMs/vApps to protect from the inventory, the
repositories with their free space, the proxies, existing credentials and encryption passwords). For a backup job
that is: name, what to protect, target repository, retention, schedule, application-aware processing and guest
credentials, encryption, and whether a backup copy or GFS is wanted; propose sensible defaults the user can accept
with one word. Then build the body from describe_endpoint / describe_model, show a short plain-language summary
of what will be created (not JSON), and call the tool only after the user agrees.
Before any action, resolve ids with read tools; never guess or construct an id. Edits (update_object,
update_settings) take only the fields to change and merge them into the current object; lists are replaced as a
whole, so send the complete new list. When you need a request body you do not know, call describe_endpoint (and
describe_model for nested or polymorphic parts, see x-subtypes) instead of guessing field names. api_write is
the escape hatch for any write endpoint without a dedicated tool. When VBR refuses a call, report its reason and do
not repeat the same call unchanged. A question about an action ("what happens if...") is not a request to run it.
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 |
|---|---|
| set_protection_group_enabledB | Enable or disable a protection group. A disabled group stops discovery and agent deployment for its computers (disabling is confirmed by the server). |
| rescan_protection_groupB | Rescan a protection group now: discover new computers and refresh agent state. Returns a session. |
| manage_discovered_computersA | Act on computers a protection group discovered (ids from vbr-mcp Additive actions run directly: install/upgrade agent, install CBT driver, rescan, trust, prepare for remote bare-metal recovery, recreate embedded recovery media. Confirmed by the server: uninstall agent / CBT driver / all components, reboot, remove (removes one computer from the group - pass exactly one id). |
| set_agent_backup_attachmentB | Attach existing backups to a managed agent computer (e.g. after reinstalling it), or detach them. |
| reboot_recovery_appliancesB | Reboot Veeam Agent recovery appliances (one by id, or all when |
| delete_backupA | Delete a whole backup (all restore points of every machine in it) from disk.
|
| delete_backup_objectA | Delete one machine's restore points from a backup ("Delete from disk" for a single VM/computer).
|
| move_backupA | Move a backup (whole chain) to another repository; the job is re-pointed to the target. Runs as a move session and can take long; jobs using the backup cannot run meanwhile. If it stops halfway,
|
| copy_backupsA | Copy backups to another repository (or, for machine backups, a folder) - the source stays as it is.
|
| move_backup_objects_to_jobB | Re-home machines' backups from one job's backup to another job (e.g. after moving VMs between jobs), so
the target job continues their chains. Without |
| run_backup_health_checkB | Run a storage-level health check (CRC/consistency) on backups, optionally repairing what it finds. Give |
| set_backup_file_malware_stateA | Mark one backup file (restore point file) as Clean or Infected for malware detection. Marking |
| manage_move_copy_sessionA | Resolve a backup move/copy session that is waiting for a decision: |
| create_recovery_mediaC | Create Veeam Recovery Media (bare-metal recovery ISO) for an agent backup object. |
| describe_endpointA | Look up the VBR REST API 1.3-rev2 reference bundled with this server. With a path and |
| describe_modelB | Schema of one VBR API model by name (e.g. Shows fields, which are required, enum values and descriptions; |
| api_readA | Send a GET request to the VBR REST API (read-only), e.g. to find ids or check state after an action.
|
| get_session_resultA | State and result of a VBR session - the asynchronous run started by almost every action tool. With |
| api_writeA | Send any state-changing request to the VBR REST API - for write endpoints no dedicated tool covers. Examples: High Availability cluster switchover, Azure / file-share / FCD Instant Recovery, Entra ID restores,
disk publishing (Data Integration API), helper appliances, KMS servers. Look the endpoint up with
|
| rescan_managed_serversB | Rescan several managed servers at once, or all of them when |
| rescan_inventoryA | Refresh the cached virtualization inventory (VMs, vApps, datastores, networks) of one managed server, e.g.
after a VM or vApp was created and does not show up in |
| set_default_mount_serverC | Make a mount server the default one for its repositories' restores (FLR, Instant Recovery). |
| get_server_certificateA | Fetch a server's TLS certificate thumbprint or SSH fingerprint - needed in the spec when adding a managed
server with |
| discover_cloud_director_vcentersA | List the vCenter Servers attached to a VMware Cloud Director server, before adding it
( |
| clone_jobA | Clone a job. VBR creates the copy under a derived name; rename or adjust it afterwards with
Supported job types: VSphereBackup, HyperVBackup, CloudDirectorBackup, BackupCopy, WindowsAgentBackup, LinuxAgentBackup. Returns the new job's model. |
| start_quick_backupA | Quick Backup: an on-demand incremental of one machine that is already protected by a backup job.
|
| apply_job_configurationB | Push a Veeam Agent backup policy's current configuration to the agents it manages ("Apply configuration"). Only for agent backup policies ( |
| clear_job_cacheA | Clear the backup cache of a Veeam Agent backup policy (the local cache agents write to while the repository is unreachable). Make sure cached restore points were uploaded first. |
| install_licenseB | Install a license file on the backup server, replacing the installed license. Confirmed by the server. |
| refresh_licenseA |
|
| set_license_optionsA | Switch license options: automatic license update (with proactive support) and instance consumption by unlicensed agents. Only the options you pass are changed. |
| manage_instance_licenseB | Assign an instance license to a workload as Server or Workstation, or |
| remove_installed_licenseB | Remove the installed license (or only its Instance or Socket section). Jobs stop processing unlicensed workloads. Confirmed by the server. |
| mark_backup_objects_cleanA | Mark backup objects (machines) as clean after a malware investigation - typically a false positive.
|
| manage_malware_exclusionsA | Create/update ( |
| start_encryption_analysisB | Run an encryption analysis for a malware event (checks whether data in the backup was encrypted, e.g. by
ransomware). Event ids: vbr-mcp |
| report_malware_eventB | Create a malware event from an external source (EDR/XDR/SIEM): VBR then flags the machine's restore points.
|
| set_ioc_monitoringA | Enable or disable monitoring of specific Indicators of Compromise (IoC) in file-level malware detection.
|
| create_objectA | Create a VBR object from a full Kinds: |
| update_objectA | Edit an existing VBR object: pass only the fields to change in Kinds: |
| delete_objectA | Remove a VBR object. Kinds: |
| update_settingsA | Edit server-wide VBR settings (or a per-object settings block) with patch semantics: pass only the fields to change; the tool reads the current settings, merges and sends the full model, and the confirmation plan shows the diff. Lists are replaced as a whole. Areas: |
| list_replicasA | List VM replicas, or - with |
| failover_replicasA | VMware vSphere replica failover. Pass one replica point (restore point of a replica) per VM.
|
| failback_replicasA | Failback of failed-over vSphere replicas, in steps:
|
| rescan_replicasB | Rescan replicas to refresh their state in VBR (all replicas when |
| restore_cloud_director_vmA | Full restore of a VMware Cloud Director VM from a restore point of a Cloud Director backup job.
|
| restore_hyperv_vmA | Full restore of a Microsoft Hyper-V VM. |
| start_instant_recovery_hyperv_vmA | Instant Recovery to Microsoft Hyper-V: run the VM straight from the backup, without a full data copy.
|
| finish_instant_recovery_hyperv_vmA | Finish a Hyper-V Instant Recovery mount: |
| start_file_level_restoreA | Mount a machine restore point for guest file-level restore (FLR). Returns the mount session; its id is the
Mounting changes nothing in production. Linux FLR uses a helper appliance/host configured in VBR. |
| browse_file_level_restoreC | List a folder (or search) in a file-level-restore mount. Read-only. |
| restore_filesA | Restore files/folders from a file-level-restore mount.
|
| unmount_file_level_restoreB | Unmount a file-level-restore mount when you are done with it. |
| change_credentials_secretA | Rotate the secret of a stored credentials record used by VBR to reach servers, guests or cloud accounts.
|
| change_encryption_passwordA | Change an encryption password (and its hint). Jobs using it encrypt new backups with the new password; existing backups still need the old one to restore - keep it safe. |
| verify_encryption_passwordA | Check whether |
| manage_user_securityB | Reset a user's MFA (they must enroll again at next login), or switch the service-account mode (service
accounts are exempt from MFA - use for automation accounts only). Roles: |
| clone_roleB | Clone a custom role; adjust the copy with |
| security_analyzerB | Security & Compliance Analyzer: |
| stop_sessionA | Stop a running session of any type: a restore, Instant Recovery, rescan, health check, SureBackup, malware
scan, move/copy, etc. (for a backup job run, vbr-mcp What is already done is not rolled back; a stopped restore can leave a partially restored VM or files behind. |
| stop_taskC | Stop one task ( |
| stop_automation_sessionB | Stop an automation (import/export) session started by |
| export_configurationA | Export VBR configuration objects as JSON import specs - for documentation, review, migration to another
VBR server, or as a template for |
| import_configurationA | Import configuration objects (as produced by |
| export_support_logsB | Collect a support log bundle (for a Veeam support case) for some jobs or hosts. With |
| send_test_emailB | Send a test message with the configured email notification settings ( |
| high_availability_clusterA | Backup server High Availability cluster: planned |
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 68 tools
Most tools target clearly distinct resource+action pairs (delete_backup vs delete_backup_object vs delete_object, move_backup vs copy_backups vs move_backup_objects_to_job, start/finish Instant Recovery pairs). A few boundaries blur: stop_session vs stop_task vs stop_automation_session, the generic create_object/update_object/delete_object family vs the dedicated create_* tools, and api_write vs any dedicated write tool. Descriptions largely resolve these by explicitly stating scope (api_write is 'for write endpoints no dedicated tool covers').
All names are snake_case, but verb conventions vary widely (set_, manage_, rescan_, start_, apply_, mark_, clone_, verify_) and a few tools are noun-only or noun-first (api_read, api_write, security_analyzer, high_availability_cluster, describe_endpoint). Patterns are still readable and mostly verb_noun, but not a single predictable scheme across 68 tools.
68 tools is far beyond a well-scoped set, driven by exposing granular operations for a very large VBR domain. While many tools are individually justified, the count creates a heavy selection burden, and generic escape hatches (api_write, create_object/update_object) coexist with numerous dedicated variants, which is exactly the strain that pushes this dimension down.
Coverage is broad: jobs, repositories, protection groups, backup deletion/move/copy, multiple restore types, malware/IoC handling, licensing, settings, security, sessions, HA and config import/export. Minor gaps remain, notably no dedicated vSphere VM full restore (only Cloud Director and Hyper-V restores) and no dedicated job-run tool (only quick backup), but api_write and companion vbr-mcp tools provide workarounds.