Skip to main content
Glama

Server Configuration

Describes the environment variables required to run the server.

NameRequiredDescriptionDefault
PROXIMO_AUDIT_KEYEDNoDefaults to on; set to 'off' to disable HMAC-SHA256 keying of the audit ledger.1
PROXIMO_ENABLE_EXECNoOpt-in: enables near-root LXC exec via SSH. Off by default.0
PROXIMO_CT_ALLOWLISTNoComma-separated list of CTIDs allowed for exec (fail-closed).
PROXIMO_ENABLE_AGENTNoOpt-in: enables qemu-guest-agent operations inside VMs. Off by default.0
PROXIMO_LEDGER_REDACTNoIf set to '1', record fingerprint instead of SQL body/command argv in ledger (privacy).0
PROXIMO_A2A_TOKEN_FILENoPath to a file containing the bearer token for A2A server (non-localhost bind requires this).
PROXIMO_AGENT_ALLOWLISTNoComma-separated list of VMIDs allowed for agent operations (fail-closed).
PROXIMO_AUDIT_EXPECTED_HEADNoExpected head hash for audit verification (off-box anchor against tail truncation/forgery).

Instructions

Guidance the server publishes about itself, which clients place ahead of the tool catalog so the model reads it before choosing anything.

This server publishes no instructions, or was last inspected before Glama recorded them.

Capabilities

Features and capabilities supported by this server

Protocol revision2025-11-25

CapabilityDetails
tools
{
  "listChanged": false
}
prompts
{
  "listChanged": false
}
resources
{
  "subscribe": false,
  "listChanged": false
}
experimental
{}

Tools

Functions exposed to the LLM to take actions

NameDescription
audit_entriesA

READ-ONLY: WHO changed WHAT and WHEN — guest configuration changes and every other audited action, read back from the PROVE ledger.

Newest first. This is how you answer "who changed this guest" or "what has this caller done". matched counts entries passing your filters, total counts the whole ledger, and truncated says so when limit cut rows. An entry with no principal returns null plus a note: the ledger not capturing an identity is a fact about the log, never a claim that nobody was responsible. This READS the chain; audit_verify PROVES it is intact.

audit_verifyA

Verify the tamper-evident audit ledger's hash chain — PROVE the log is intact.

Pass expected_head (the head() value you pinned off-box) to also catch tail truncation, a forged tail-append, or a full file replacement — a forward walk alone can't see those. Falls back to PROXIMO_AUDIT_EXPECTED_HEAD when omitted.

proximo_callA

Call any Proximo tool by exact name, including ones not in this server's listed tools.

Get the argument shape from proximo_tool_schema first. Same gates as calling it directly: dry-run PLAN, ledger entry, token ACL. A smaller doorway, not a looser one.

proximo_recallA

READ-ONLY: the estate map from local Tier-1 memory — NOT a live PVE read. Returns total/by_kind/by_status/guest_summary counts (trust guest_summary for guest-count questions; all counting is server-side) plus lean entity rows, stamped {source:'memory', as_of, age_seconds}: the data is as old as the stamp says. With since, also diffs: appeared, status_changed, and not_seen_since (last observed before the window — a fact, not a claim the entity is gone). journal=N adds the newest N diagnosis digests ("when did this last happen") — findings summaries only, never raw diagnostic output. Memory is on by default (PROXIMO_MEMORY=0 opts out), fed opportunistically by list reads and diagnose/doctor runs, derived and rebuildable. For live state use pve_list_guests / pve_cluster_resources.

proximo_find_toolsA

Search Proximo's full tool catalog by keyword.

ESTATE QUESTIONS FIRST: if the question is what exists, how many, what changed, or when something last happened, call proximo_recall instead — it answers from local memory in one call, with no search and no schema lookup, and it stamps how old the answer is. Come here for everything else.

The facade is resident; 908 more tools on this server are searchable but not. Search for what you want ("guest power", "ceph pool", "firewall"), then call proximo_tool_schema on a result to get its arguments, then proximo_read (read-only tools) or proximo_call to run it. All terms must match.

proximo_tool_schemaB

Full description + JSON input schema for one tool found via proximo_find_tools.

proximo_readA

READ-ONLY: run a read-only Proximo tool by exact name; refuses anything that can mutate (use proximo_call for those). Same flow: get the shape from proximo_tool_schema first.

Prompts

Interactive templates invoked by user choice

NameDescription
safe_migrationRunbook: migrate a guest to another node safely (plan-first, verify-after).
diagnose_clusterRunbook: read-only health sweep of the cluster (DIAGNOSE, no changes).
provision_containerRunbook: provision a new LXC within policy (plan-first, verify-after).
safe_backupRunbook: back up a guest and verify the backup actually landed.
review_receiptsRunbook: verify Proximo's PROVE ledger integrity (the receipts).

Resources

Contextual data attached and managed by the client

NameDescription

No resources

TDQS

A4.1/5.0

Scored across 7 tools

Disambiguation4/5

Each tool has a distinct role: find/schema/read/call form a discovery-execution pipeline, recall serves estate/memory questions, and audit_entries/audit_verify handle ledger inspection and integrity. The only mild overlap is between proximo_recall and audit_entries for "what changed" questions, and between proximo_read and the broader proximo_call, but the descriptions draw clear lines.

Naming Consistency4/5

Names use consistent snake_case and clear domain prefixes (proximo_ vs audit_), which makes grouping obvious. However, not all names follow a single verb_noun pattern: proximo_tool_schema and audit_entries are noun-style while the rest are verb-led, so it is slightly mixed rather than fully uniform.

Tool Count5/5

Seven exposed tools is appropriate for a facade server: search, schema lookup, read-only execution, general execution, memory recall, and audit read/verify. Each earns its place, and the 908 underlying tools are intentionally indexed behind find_tools instead of being exposed directly.

Completeness4/5

The discovery → schema → execute lifecycle is fully covered, with audit read/verify and estate recall rounding out the domain. A full catalog listing tool is absent, but keyword search plus exact-name calling covers the practical workflow, so there are no serious dead ends.

Maintenance

ActivityActive
ResponsivenessResponsive