Repository navigation
Releases
ProxmoxMCP-Plus v0.7.1
ProxmoxMCP-Plus v0.7.1
- Update Code Mode to Monty 1.6.0 and lock the matching Python client and
worker distributions with hashes for CI and container builds.
#212 - Pin the CodeQL initialization and analysis steps to the same reviewed
upstream revision.
#213
Install proxmox-mcp-plus==0.7.1 or pull
ghcr.io/rekklesna/proxmoxmcp-plus:0.7.1, then restart MCP workers. Existing
configuration and job databases can be reused without migration.
ProxmoxMCP-Plus v0.7.0
ProxmoxMCP-Plus v0.7.0
- Require 100% statement and branch coverage in every runtime module, with
exact counts checked independently on Python 3.11 and 3.12. - Keep stopped-container memory usage at zero when historical RRD data still
contains samples from before shutdown. - Require OpenTelemetry API 1.45.2 or later in Python installations and pin
1.45.2 with hashes in the container runtime lock. This addresses
GHSA-mf5v-w3w3-f6cg. - Save Proxmox task status when the task log cannot be read. The last observed
progress remains available, and the job audit recordspolled_without_log. - Integrate #208 to
advertise idempotency for read-only tools and Code Mode lookups. Target
discovery advertises network access and retries after malformed inventory
without caching it. - Measure OpenAPI rate-limit windows with elapsed time so wall-clock corrections
do not freeze or clear a client's request allowance. - Copy only packaging sources, runtime dependencies and example configurations
into Docker images. CI verifies that an unrelated build-context file stays out
of the image and that the runtime user cannot write directly to the source directory.
Upgrade to proxmox-mcp-plus==0.7.0 or pull
ghcr.io/rekklesna/proxmoxmcp-plus:0.7.0. Recreate containers, restart
workers and reconnect MCP clients to refresh tool metadata. Existing job
databases and configuration can be reused without a migration.
ProxmoxMCP-Plus v0.6.10
ProxmoxMCP-Plus v0.6.10
Changes
- Upgrade Monty, its client and runtime packages from 1.4.0 to 1.5.0 for Code Mode (#207). Refresh the universal runtime lock and hashes so source installs and container images use the same versions.
- Update the pinned GHCR login action from 4.7.0 to 4.8.0 (#200).
- Update CodeQL init and analyze together to the same pinned revision (#201, #202, #206), resolving their version mismatch.
- Group future CodeQL action updates in Dependabot to keep both steps aligned (#203).
Upgrade
Install proxmox-mcp-plus==0.6.10 or pull ghcr.io/rekklesna/proxmoxmcp-plus:0.6.10. Restart workers or recreate containers and reconnect MCP clients. No configuration or database migration is required. Preserve existing configuration, credentials and job volumes.
To roll back, reinstall 0.6.9 or use its container image. Rolling back cannot reverse Proxmox operations already performed.
Validation
The release gates cover Python 3.11 and 3.12, CodeQL, dependency and secret scanning, Windows contracts, container health and authentication, and package metadata. No live Proxmox operations are required by CI.
ProxmoxMCP-Plus v0.6.9
ProxmoxMCP-Plus v0.6.9
Changes
- Upgrade pydantic-monty, pydantic-monty-client and pydantic-monty-runtime
from 1.2.0 to 1.4.0 for Code Mode (#196). Regenerate the universal runtime
lock and package hashes so source installs and container images use the
same dependency versions. - Allow mypy 2.x in development dependencies (#197).
- Update the SHA-pinned docker/login-action to 4.7.0 for GHCR publishing (#198).
Upgrade
Install proxmox-mcp-plus==0.6.9 or pull
ghcr.io/rekklesna/proxmoxmcp-plus:0.6.9. Restart workers or recreate containers
and reconnect MCP clients after upgrading. No database migration or new
configuration is required. Preserve existing credentials, configuration and job
volumes. To roll back, reinstall 0.6.8 or use its container image; this does not
reverse Proxmox operations already performed.
Validation
Release gates cover Python 3.11/3.12 tests and coverage, Code Mode behavior,
Windows contracts, container health and authentication, static analysis,
dependency audits and package metadata consistency. Validation uses mocked
Proxmox APIs; no live guest operations are performed.
ProxmoxMCP-Plus v0.6.8
ProxmoxMCP-Plus v0.6.8
Changes
- Advertise MCP
readOnlyHint,destructiveHint,idempotentHintand
openWorldHintannotations for all 65 built-in tools. Read-only queries can be
distinguished from mutations by clients that support these hints. - Correct classifications from #194:
container resource updates are non-idempotent because repeateddisk_gbcalls
keep growing the disk; backup creation is potentially destructive because
mode=stopormode=suspendinterrupts guests. Guest starts are conservatively
destructive because they change power state and launch workloads. - Annotate Code Mode's three public tools: local search and schema lookup are
read-only and closed-world; execution is potentially destructive,
non-idempotent and open-world because it can invoke mutating domain tools. - Keep explicit plugin annotations and leave unknown extensions unclassified.
Target discovery and job listing are local; job observation tools may update
local tracking/audit data but do not submit Proxmox mutations.
Annotations are advisory metadata, not authorization or automatic retry
permission. Existing read-only target enforcement, client permissions, command
policy and approval checks remain in effect. Clients must establish trust in a
server before relying on its annotations.
Upgrade
Install proxmox-mcp-plus==0.6.8 or pull
ghcr.io/rekklesna/proxmoxmcp-plus:0.6.8. Restart workers, recreate containers
using the new image and reconnect MCP clients to refresh tools/list.
No database migration or new configuration is required. Keep existing
configuration, credentials and job volumes. Downgrading to 0.6.7 removes the new
annotations; it does not reverse guest or storage operations already performed.
Validation
Regression tests demonstrate repeated disk growth and backup stop/suspend
forwarding with a mocked Proxmox API. Tests cover all registered annotations,
explicit overrides, unknown extensions, stdio discovery and Code Mode protocol
discovery. Release gates also check Python 3.11/3.12, complete runtime statement
coverage and critical-module branch coverage, Windows contracts, container
health/authentication, static analysis, dependency audits and built packages.
No live Proxmox guest operations are used for release validation.
ProxmoxMCP-Plus v0.6.7
ProxmoxMCP-Plus v0.6.7
Changes
- Add optional
purge=falsetodelete_vmanddelete_container. Explicitly
settingpurge=truealso removes HA resources and backup/replication references
through the native Proxmox delete endpoint. Preserve the selected option across
in-process and persisted retries, including after a job-store restart. Older
retry recipes without the field retain their previous behavior. - Fix forced deletion of HA-managed guests: after the stop task succeeds, poll
the guest's current status for up to 60 seconds and submit deletion only after
it reportsstopped. A timeout or failed status read prevents the delete
request; the earlier stop request can still complete later. - Preserve existing positional approval arguments for Python callers. The new
purgeargument is keyword-only, preventing an approval token from accidentally
enabling additional cleanup. Existing target, read-only, and approval policies
continue to apply. - Update Monty, its client, and runtime to 1.2.0 with a synchronized universal
hashed lock. Keep the existing Code Mode restrictions and execution limits.
Add a dependency-lock regression check and installed-dependency checks to CI
and Docker builds. - Update certifi to 2026.10.19 to address
GHSA-6q48-6c7w-9gw2.
Upgrade SHA-pinned setup-python to 7.1.0 and the PyPI publishing action to
1.14.3; allow Paramiko 5 typing stubs. - Add contribution guidelines, Contributor Covenant 2.1, bug/feature/accessibility
issue forms, a PR template, and an accessibility statement. Align documented
validation commands with the current CI requirements.
These changes integrate #185
through #186,
#187, and dependency PRs
#188,
#189,
#190, and
#191 through
#192.
Upgrade
Install proxmox-mcp-plus==0.6.7 or update the container image to
ghcr.io/rekklesna/proxmoxmcp-plus:0.6.7, then restart workers and reconnect MCP
clients to refresh tool schemas. Recreate Docker containers after pulling the new
image. Keep the existing configuration and persistent jobs volume.
No database schema migration or new configuration setting is required. Existing
deletion calls keep purge=false. For direct Python calls, opt in explicitly with
purge=True; named MCP arguments use purge=true. Proxmox removes guest ACLs
during normal deletion regardless of this flag. See the
API reference for deletion parameters.
Deleting a guest and purging its references cannot be undone by downgrading the
software. Restore guest data from backups and restore cluster references
separately if recovery is needed. A software rollback to 0.6.6 loses the new purge
option and HA stop-state wait, and returns to the older dependency set.
Validation
The integrated changes passed 1,433 tests with 2 optional skips on both Python
3.11 and 3.12 in CI, including complete runtime statement coverage and complete
branch coverage of the nine critical modules. The release PR runs these gates
again, alongside container health/authentication, Windows contracts, dependency
audits, package checks, secret scanning, and CodeQL. No live Proxmox guest is
stopped or deleted as part of release validation.
ProxmoxMCP-Plus v0.6.6
ProxmoxMCP-Plus v0.6.6
Changes
- Fix #183: default Compose host publications to IPv4 loopback for native MCP HTTP (
8000) and OpenAPI (8811).MCP_HTTP_BIND_ADDRESSandOPENAPI_BIND_ADDRESSprovide explicit per-mode overrides. - Put OpenAPI behind the
openapiprofile; both Compose modes are opt-in. Remove the cross-mode required-variable interpolation so MCP-only deployments no longer needPROXMOX_API_KEY. Application startup continues to enforce the selected mode's authentication. - Enable
no-new-privilegesand drop all Linux capabilities on both example services. Verify hardened OpenAPI/native HTTP operation and missing-credential rejection in container CI, with native ARM release health checks under the same capability restrictions. - Document profile selection, reverse proxies/tunnels, container-network routing, explicit LAN bindings and migration from the previous default OpenAPI service.
Upgrade
Update/rebuild the image and Compose file, set the selected mode's credential, and recreate the wanted service:
docker compose --profile openapi up -d --build --force-recreate proxmox-mcp-api
# Or use the native MCP HTTP mode:
docker compose --profile mcp-http up -d --build --force-recreate proxmox-mcp-httpPlain docker compose up -d now selects no services. Existing containers keep their old configuration until recreated. If migrating to MCP-only mode, explicitly stop an old OpenAPI service with docker compose --profile openapi stop proxmox-mcp-api. The named jobs volume is preserved.
Host-side proxies/tunnels can use loopback. Containerized proxies use the shared Docker network and service DNS. For direct remote access, explicitly set the appropriate bind-address variable; container listeners and the bare image default mode remain unchanged. See Compose deployment.
Install proxmox-mcp-plus==0.6.6 or ghcr.io/rekklesna/proxmoxmcp-plus:0.6.6. Python tool behavior and database schemas are unchanged. No production Proxmox operation is used for validation.
ProxmoxMCP-Plus v0.6.5
ProxmoxMCP-Plus v0.6.5
Changes
- Integrate #181: add
download_templatefor HTTPS LXC archives through the native Proxmoxdownload-urlAPI withcontent=vztmpland TLS verification enabled. Validate identifiers, archive names and optional digests; require observed enabled/active template storage and reject an existing destination volume. - Apply target-specific high-risk approval, read-only and client-grant policy to the new tool. Accept approval metadata without forwarding it to Proxmox.
- Report submission separately from completion, preserve UPIDs if local tracking fails, and classify uncertain responses as unsuccessful/non-retryable. Do not attach a blind retry recipe. Use the native task DELETE endpoint for cancellation.
- Integrate #178: permit Ruff 0.17 while explicitly preserving the existing
E4/E7/E9/Flint baseline and Python 3.11 target. Review adoption of new rule families independently from a dependency update. - Add a pull-request AMD64 container build/health job so Dockerfile updates are exercised before merging, without publishing a PR image.
- Integrate #179: update the pinned official Python 3.11 slim base-image digest; retain Linux AMD64/ARM64 publication and native ARM runtime checks.
Upgrade and operating limits
Install proxmox-mcp-plus==0.6.5 or ghcr.io/rekklesna/proxmoxmcp-plus:0.6.5, restart workers and reconnect MCP clients. No state/database migration is introduced. Add download_template to any custom command_policy.high_risk_operations list. Default high-risk mode is audit-only; configure enforcement and approval tokens when required.
Use trusted publishers and preferably SHA-256/SHA-512 digests. Proxmox performs the download from its own network, so HTTPS validation does not prevent access to internal endpoints or redirects. Proxmox enforces storage/network permissions, TLS, checksum verification and task failures.
Inventory checks do not reserve disk space or the filename, and cannot prevent an external concurrent upload. Downloads can fail asynchronously. Poll the task and verify list_templates before using the resulting volume with create_container. After SUBMISSION_UNCERTAIN, inspect tasks and storage before another call. retry_job deliberately does not resubmit template downloads.
Validation uses automated tests, simulated API/protocol calls, package installation and container smoke checks. No production Proxmox download or guest operation is performed.
ProxmoxMCP-Plus v0.6.4
ProxmoxMCP-Plus v0.6.4
Changes
- Add
preflight_vm_migrationandmigrate_vmfor offline QEMU migration. Read native Proxmox migration preconditions alongside VM ownership/runtime, HA, storage and network evidence; reject missing or malformed evidence, local ISO media and unsupported host/device dependencies. - Preserve shared-storage semantics and generated CloudInit media. Use the larger configuration/native disk size for local capacity accounting, without charging shared disks their existing provisioned space again.
- Add
import_cloud_imagefor already uploaded raw/qcow2/vmdk volumes on import-enabled storage. Submit the asynchronous Proxmoximport-fromupdate with the current config digest into an unused SCSI slot, without URL downloads, disk overwrites or boot-order changes. - Add idempotent
configure_vm_serial_consolefor socket-backed ttyS0 and serial VGA output while preserving other serial devices. Guest serial login requires separate guest configuration. - Repair the contributed tools' approval-token dispatch, template/path validation, digest checks and uncertain submission envelopes. Enforce the selected target's high-risk/read-only/client policy, never retry mutation requests automatically, and retain a submitted UPID if local job tracking fails.
- Integrate contributions from #176 and #177 with independent API review and regression coverage of real FastMCP dispatch, negative authorization, native migration evidence and mutation outcomes.
Upgrade and operating limits
Upgrade to proxmox-mcp-plus==0.6.4 or ghcr.io/rekklesna/proxmoxmcp-plus:0.6.4, restart workers and reconnect MCP clients. Preserve existing SQLite/OAuth state; no database migration is introduced.
Custom command_policy.high_risk_operations lists must include migrate_vm, import_cloud_image and configure_vm_serial_console to apply their approval policy to these tools. Default audit-only mode remains available; explicit tool confirmations are still required.
QEMU migration is offline only: stop the VM separately and verify backup restorability, workload drain and destination CPU compatibility before confirming. The native migration GET requires VM.Migrate permission even though the preflight operation is read-only. Unsupported/missing native preconditions fail closed. Migration observations neither reserve resources nor atomically lock runtime state; Proxmox enforces its own submission constraints.
Disk import and serial setup also observe stopped state before submission and use a config digest against concurrent config edits; the digest does not lock external runtime changes. Inspect task completion and resulting configuration. On SUBMISSION_UNCERTAIN, inspect Proxmox tasks, VM config and guest location before considering another call.
No live Proxmox VM migration, image import or guest configuration change was performed during this code review. Validation uses automated unit/protocol tests and publication/runtime smoke checks.
ProxmoxMCP-Plus v0.6.3
ProxmoxMCP-Plus v0.6.3
Changes
- Add
get_node_timefor node clock and timezone inspection,get_node_storagefor capacity on the selected node, andget_node_diagnosticsfor CPU, memory, storage and observed RRD I/O metrics. Missing optional metrics produce partial results instead of a false completeness claim. - Add
get_cluster_join_info, exposing documented public cluster metadata and certificate fingerprints while omitting credentials and unrecognized extensions. - Add read-only
preflight_container_migrationand guardedmigrate_container. Both use fresh checks for node availability, guest state, rootfs storage and capacity, mounts/devices, HA membership and matching backup inventory. Missing or malformed evidence blocks readiness. - Submit migration with LXC-supported API parameters and explicit migration/downtime confirmations. Apply the selected target's configured high-risk policy, return the submitted task UPID and optional local job ID, and never automatically retry migration. Ambiguous transport or submission responses return
SUBMISSION_UNCERTAIN. - Keep reachable-node backup results when other cluster members are offline, and report an incomplete inventory instead of a false empty result.
- Align the new tools across the manifest, canonical catalog, read-only policy and operator documentation. Preserve contributor attribution from #167–#172 through the reviewed integration in #174; the offline-backup fix is #173.
Validation
The included runtime changes passed GitHub CI on Python 3.11 and 3.12 with PostgreSQL: 1,043 tests passed and 2 optional integration tests skipped on each version. All 61 runtime modules have zero uncovered statements and all 9 critical modules have zero uncovered branches. Windows contracts, Ruff, Mypy, package checks, dependency audits, secret scanning and CodeQL passed. The release metadata update follows the same required CI gates before publication.
No live Proxmox migration was performed. Migration assessment is advisory: it does not reserve capacity or verify backup freshness/restorability, application health or every Proxmox migration constraint. Complex mounts, unused volumes, custom devices and HA-managed guests require separate operator handling.
Upgrade
Upgrade to proxmox-mcp-plus==0.6.3 or ghcr.io/rekklesna/proxmoxmcp-plus:0.6.3, restart workers, and reconnect MCP clients to refresh tool discovery. Preserve existing SQLite state and legacy OAuth metadata; no database migration is introduced.
The default high-risk operation list now includes migrate_container. If you supply a custom command_policy.high_risk_operations list, add this tool explicitly to apply your configured approval policy. confirm_migration=true is required for every migration; restart=true also requires confirm_downtime=true. Default audit-only mode remains available, and read-only targets reject migration submission.
After submission, poll the job/task and inspect the destination guest. If SUBMISSION_UNCERTAIN is returned, inspect Proxmox tasks and guest location before considering another call.