Skip to main content
Glama

Server Configuration

Describes the environment variables required to run the server.

NameRequiredDescriptionDefault
COLMENA_HOMENoDirectory where colmena state is stored (jobs/, worktrees/, reports/, usage.json, activity.log, audit.log). Default is ~/.colmena/.
COLMENA_CONFIGNoPath to the colmena.yml configuration file. Default is ~/.colmena/colmena.yml.
COLMENA_NOTIFIERNoNotifier type: toast, log, or none.
COLMENA_CODEX_BINNoPath to the Codex CLI binary. Auto-detected if not set.
COLMENA_MAX_PARALLELNoMaximum number of parallel agents (default from colmena.yml defaults.max_parallel, e.g. 4).
COLMENA_OPENCODE_BINNoPath to the OpenCode binary. Auto-detected if not set.
COLMENA_JOB_TIMEOUT_MINNoJob timeout in minutes (default from colmena.yml defaults.timeout_min, e.g. 60).

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
}

Tools

Functions exposed to the LLM to take actions

NameDescription
workersA

Lista los workers de colmena.yml con prioridad, fortalezas, gasto del mes, presupuesto restante y disponibilidad. Úsalo para decidir a quién delegar.

reportA

Genera un reporte HTML bonito de lo logrado por los jobs indicados (ronda final). El worker del rol reporter lo redacta con el estilo elegido (defecto papercraft); si falla, se usa un renderizador de respaldo. Devuelve la ruta del HTML.

runB

Lanza un agente obrero (Codex u OpenCode según el worker; sin aprobaciones, con internet). Devuelve job_id; con wait=true espera el reporte (hasta timeout_s).

run_manyB

Lanza varios agentes en paralelo (tope COLMENA_MAX_PARALLEL simultáneos, el resto en cola).

handoffC

Lanza un agente nuevo con el resumen de otros agentes como contexto.

statusC

Estado de un job.

listC

Lista jobs recientes.

waitB

Espera a uno o todos los jobs (máx 600 s) y devuelve reportes compactos.

resultC

Reporte compacto (RESULT/FILES/TESTS/RISKS), archivos tocados y rama.

diffC

Diff del trabajo (truncado; el completo queda en disco).

continueC

Reanuda la sesión de un agente con un mensaje de seguimiento (mismo workdir).

summarizeC

Un agente (rol summarizer) resume la sesión de un job.

compactC

Comprime la sesión de un job en un brief de relevo (summary.md) reutilizable con handoff.

reviewC

Revisión en solo lectura de los cambios de un job (o de un cwd).

cancelB

Cancela un job (mata el árbol de procesos).

mergeB

Fusiona la rama colmena/ del job en la rama actual de su cwd. Manual: revisa antes con diff/review. Se niega si el destino tiene cambios sin commitear.

cleanupB

Elimina worktrees y ramas colmena/* creados por este servidor (de un job, o más viejos que N días).

Prompts

Interactive templates invoked by user choice

NameDescription

No prompts

Resources

Contextual data attached and managed by the client

NameDescription

No resources

TDQS

B3/5.0

Scored across 17 tools

Disambiguation4/5

The set clearly separates most orchestrator actions, but summarize vs compact and wait vs result share reporting/summarization duties, and run/handoff/continue form a related launch-resume family. The descriptions usually resolve these overlaps, though a few boundaries require careful reading.

Naming Consistency4/5

Names are consistently lowercase and mostly single words, mixing imperative verbs (run, cancel, merge) with resource nouns (workers, status, result) in a readable way. The only multi-word name, run_many, uses snake_case cleanly, so the convention is predictable despite the verb/noun mix.

Tool Count4/5

17 tools is slightly above the ideal 3–15 range, but the agent-orchestration domain is broad and each tool maps to a distinct lifecycle action. No tool is clearly redundant, though a leaner design could combine some summary/reporting operations.

Completeness4/5

The surface covers the orchestration lifecycle well: delegation, parallel runs, handoff, monitoring, cancellation, review, merge, cleanup, reporting, and session summarization/compaction. Minor gaps include no tools for inspecting raw job logs or configuring worker budget/priority settings, but agents can mostly work around these.

Maintenance

ActivityMaintained
ResponsivenessNo issues