Skip to main content
Glama

Server Configuration

Describes the environment variables required to run the server.

NameRequiredDescriptionDefault

No arguments

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": true
}
prompts
{
  "listChanged": false
}
resources
{
  "subscribe": false,
  "listChanged": false
}
experimental
{}

Tools

Functions exposed to the LLM to take actions

NameDescription
start_agentA

Start a private CLI task; invocation_id goes to wait, not session_id. No deadline or messages to active tasks; cancel to stop. Reply settings are requested, not effective; unreceived flags undelivered ends/blockers/reviews for wait.

waitA

Return when any handle is ready; reason=ready contains all ready now in updates (tasks) and rounds, not necessarily all finished.

Ready: terminal; execution_blocked (still running; once/category); review_needed (pending_review needs decide_answers); seat_lost (failed/blocked seat while peers run; once/seat, answers withheld; retry via run_round.retry_seats). pending_ids lists unreceived ends for the given ids or inbox; all_finished means none remain. Handle blockers/reviews, then wait on pending_ids until all_finished; drop delivered handles. Ends are never pushed: wait after dispatch or independent work; collect every terminal result before ending your turn. wait_elapsed gives status/activity; wait again normally. Only wait_seconds=0 adds running tails (up to 1000 characters) and age, subject to include_output. Round answers arrive once; output_returned means reread with read_conversation. Unknown/expired/duplicate ids reject the whole call. Inbox ends arrive once; reviews stay ready until decided; nothing_pending means an empty inbox.

cancelA

Stop a task or round; return its cancelled or already-finished result. Running seats become absent, unstarted seats skip, published answers stay and undecided answers stay private. CLI side effects are not undone.

listC

List invocation metadata and saved conversations, newest first; open a read-only dashboard. writable=false: continue at the owning server or when idle. Private: never send to seats.

list_membersB

Read App members, defaults, review/orchestration seats and mode limits afresh. members merges provider/model rows: omitted modes means all modes, [] means none; default is true or a list of default modes. error checks Adapter options, not installation, authentication or model availability; repair invalid files in App. Missing members leave modes unrestricted; disabled orchestration supplies no defaults. Omitted available means usable; unavailable seats may offer replacement parameters: pass with original seat/role; switching is never automatic. Native seats/exclusions: see MCP instructions.

run_roundA

Create/continue a saved meeting; group_id equals conversation_id. Seats keep native sessions across rounds/restarts and automatically share committed entries. New: title/members required; add prompt/seats to run round 1, omit prompt to only create. Existing: omit creation-only arguments. One active round per meeting; wait on round_id. entries contains committed message_ids; unreceived flags ends/blockers/reviews for wait. At meeting close, post the summary you give the user as a prompt with no seats.

decide_answersA

Decide held answers atomically: publish shares with peers; private keeps between boss and seat. Repeating a decision is idempotent; changing it is rejected. Parallel rounds commit after all decisions; serial rounds then start the next seat. Returns after local commit when no work/review remains: finished needs no further call; running/review requires wait.

handoverA

Transfer all tasks/meetings of a chat only on user request; no arguments lists waiting workspaces. release=true hands off; workspace_id claims and refuses the previous boss. Claiming your own waiting workspace withdraws it. Running work finishes on the old server; continue when writable=true. Private: never send to seats. Follow the agentnave-handover Skill.

read_conversationB

Read saved final messages/tasks, including native session_ids and forwardable message_ids. Meeting id equals group_id; pending_answer marks answers awaiting review. Private: never forward the whole result to seats.

Prompts

Interactive templates invoked by user choice

NameDescription

No prompts

Resources

Contextual data attached and managed by the client

NameDescription

No resources

TDQS

A3.5/5.0

Scored across 9 tools

Disambiguation4/5

The tools mostly target distinct orchestration objects and actions: start_agent for CLI tasks, run_round for meetings, decide_answers for held answers, and handover for workspace transfer. Some overlap exists among list, list_members, and read_conversation, and wait spans multiple handle types, but the descriptions provide enough differentiation.

Naming Consistency3/5

Most compound names follow imperative snake_case verb_noun patterns such as start_agent, list_members, run_round, decide_answers, and read_conversation. However, bare verbs wait/cancel/list and the noun-style handover break the predictable pattern, making it mixed but still readable.

Tool Count5/5

The server exposes 9 tools, which is well within a reasonable range for an orchestration service. Each tool appears to earn its place by covering a distinct lifecycle or control operation, with no obvious redundant tools.

Completeness4/5

The surface covers task start/wait/cancel/list/read, meeting round creation/continuation, answer decisions, member inspection, and handover. Minor gaps exist, such as no explicit member configuration/update tool and no messaging to active tasks, but some of these appear intentionally out of scope.

Maintenance

ActivityActive
ResponsivenessUnresponsive