Dos agentes en entornos diferentes pueden descubrirse, intercambiar mensajes y dividir el trabajo sin que un humano tenga que transmitir el contexto entre ellos:
Claude Code → Concord Claim src/app/page.tsx
Codex → Concord Claim src/app/page.tsx
Concord → Codex Overlap: Claude Code already owns this file
Codex → Claude I'll take src/app/api instead. Does that work?
Claude → Codex Yes. I'll keep the page and use your API contract.Ejecuta la demo real de Claude Code ↔ Codex para ver cómo ambos agentes resuelven una reclamación superpuesta mediante un prompt/respuesta en vivo, crean una aplicación jugable, transfieren la propiedad y entregan el resultado a un revisor independiente.
Related MCP server: ccg-mcp-tool
npm install -g @concord-ai/concord-mcp
cd /path/to/your/repository
concord setupReinicia tus clientes de agente y pide a dos de ellos que trabajen en el mismo repositorio. Concord les proporciona un espacio de trabajo compartido y hace que las sesiones alcanzables estén disponibles para prompts y respuestas directos.
concord setup crea el espacio de trabajo local .concord/, registra el servidor MCP
para Claude, Cursor, Gemini, Grok y Codex (.mcp.json, .cursor/mcp.json,
.gemini/settings.json, .grok/config.toml y ~/.codex/config.toml) y escribe
las instrucciones de las herramientas de Concord en las configuraciones de tus clientes (CLAUDE.md, AGENTS.md,
.codex/, .cursor/rules/). Se fusiona con la configuración existente en lugar de reemplazarla
y es seguro volver a ejecutarlo.
El setup también detecta clientes compatibles e intenta instalar sus adaptadores globales de Concord
de forma independiente. Usa --no-adapters para omitir ese paso o
--require-adapters en instalaciones gestionadas que deberían fallar si el soporte está degradado.
Pasa --no-mcp para escribir solo el espacio de trabajo y las instrucciones mientras gestionas
el registro de MCP por tu cuenta.
Agente | Guía de integración |
Claude Code | |
Codex | |
Cursor | |
Gemini CLI | |
Grok Build | |
Cualquier otro agente de codificación compatible con MCP | Estado de trabajo compartido a través de las cinco herramientas MCP |
No existe un comando universal
/concord— los comandos son específicos de cada cliente. Concord funciona a través de las herramientas MCP más las instrucciones instaladas en cualquier cliente compatible con MCP.
La entrega en vivo depende del entorno receptor y del estado de la sesión. Ejecuta
concord adapters status para ver qué agentes instalados son alcanzables y cómo
se entregarán los mensajes.
Los mensajes hacen que los agentes hablen. El estado de trabajo compartido de Concord mantiene la colaboración resultante fiable después de que el mensaje se entregue.
Sin Concord | Con Concord |
Los agentes no pueden contactar con pares en otro entorno | Los agentes envían prompts directos y respondibles entre clientes compatibles |
Los agentes descubren colisiones después de editar | Los agentes reclaman archivos y módulos antes de comenzar el trabajo |
El contexto desaparece cuando una sesión termina | Las decisiones, suposiciones y hallazgos permanecen vinculados a la tarea |
La propiedad se deduce del historial de chat | Las asignaciones y transferencias son explícitas y se reconocen |
Los humanos reconstruyen el progreso a partir de ramas y diffs | Los paquetes de revisión llegan con alcance, pruebas, riesgos y procedencia |
Concord no es otro agente autónomo ni orquestador. Es la capa compartida alrededor de tus agentes: presencia, mensajería, memoria de tareas, propiedad, transferencias y estado de revisión a través de un pequeño servidor MCP.
Herramienta | Propósito |
| registra la presencia, reclama o acepta una tarea y reporta solapamientos de alcance antes de editar |
| lee el estado del espacio de trabajo/tarea, la bandeja de entrada/salida de un agente o un hilo de prompt/respuesta duradero |
| registra el contexto de la tarea o envía inmediatamente un prompt/respuesta a otro agente del espacio de trabajo con prompt |
| asigna, acepta, rechaza, libera, reasigna, ofrece transferencias o reabre trabajo versionado |
| registra evidencia y opcionalmente marca una tarea como lista para revisión, completada o cerrada |
Las escrituras aceptan un agent_id, que mantiene la presencia activa simplemente trabajando.
inspect_work muestra quién está aquí y marca reclamaciones obsoletas — una
reclamación activa cuyo agente propietario se ha ido sin transferirla.
Para la comunicación en vivo entre agentes, ejecuta concord setup y luego reinicia las
sesiones de cliente existentes una vez. Un prompt usa update_work con operation: "prompt", el
to_agent_id de destino, el contenido y una idempotency_key; una respuesta usa
operation: "reply" y reply_to_message_id. Un adaptador con confirmación de entrega dirige
un turno ocupado o inicia un turno inactivo. Las integraciones solo con hooks dejan un mensaje
de extracción duradero y limitan esa limitación en el resultado. La entrega falla inmediatamente
cuando el agente nombrado no tiene un endpoint alcanzable; Concord no lo redirige silenciosamente.
concord adapters status informa de cada entorno por separado, incluyendo su
tipo de monitor/controlador, alcanzabilidad verificada, acción requerida y resultado de la
sonda de versión. concord adapters install, doctor y uninstall proporcionan el
mismo ciclo de vida global fuera de la configuración del repositorio.
Concord resuelve el espacio de trabajo del repositorio automáticamente. Las operaciones devuelven su
workspace_id y la raíz del repositorio para que un cliente pueda detectar una llamada mal enrutada; el
id se puede pasar explícitamente cuando un servidor coordina múltiples raíces.
Las operaciones que cambian el ciclo de vida usan la version monotónica de la tarea como
expected_version. Si dos agentes actúan sobre la misma versión, solo la primera
transición tiene éxito. La asignación deja el trabajo en assigned hasta que el agente nombrado
use transfer_work con action: "accept"; una oferta de transferencia también mantiene la
propiedad con el remitente hasta que el destinatario la acepta. Cada cambio de propiedad
se conserva en un historial de auditoría de solo añadir.
SQLite es la fuente de verdad local, almacenada en .concord/ en la raíz del
repositorio donde se realiza el trabajo. El servidor MCP resuelve esa raíz desde
CONCORD_REPO_ROOT si está definido, luego desde CLAUDE_PROJECT_DIR de Claude Code (que Claude Code
establece automáticamente, incluso para un servidor con ámbito de usuario), y luego desde su
directorio de trabajo — de modo que cada agente en un repositorio comparte un único almacén. Establece CONCORD_REPO_ROOT
cuando ejecutes el servidor en un lugar donde su directorio de trabajo no esté dentro del repositorio.
Los worktrees de Git vinculados siguen los metadatos commondir de Git hasta el checkout principal,
de modo que el checkout principal y todos los worktrees vinculados comparten intencionadamente una única
base de datos y un único id de espacio de trabajo de Concord.
Para restringir la selección explícita de espacios de trabajo, establece CONCORD_ALLOWED_ROOTS a una
lista de raíces de repositorio permitidas separadas por rutas. Sin una lista de permitidos, las raíces
decodificadas deben existir y ser directorios.
concord setup añade .concord/ al
.gitignore del repositorio, de modo que el espacio de trabajo generado permanezca local por defecto.
Los equipos que quieran artefactos seleccionados en los PRs pueden eliminar esa regla o forzar la adición de los
archivos legibles por humanos:
.concord/
├── concord.db local source of truth
├── HANDOFF.md human-readable handoff
├── REVIEW_PACKET.md review-ready evidence
└── WORK_STATE.json generated export (optional)Concord admite tanto herramientas MCP tipadas como una CLI normal. Los agentes compatibles con MCP pueden
llamar a las herramientas directamente; los humanos y los agentes orientados a CLI pueden trabajar con el mismo
espacio de trabajo compartido a través de los comandos concord.
concord setup # set up local state, instructions, and MCP clients
concord status # roster, active work, overlaps, stale claims, review-ready
concord dashboard # live, keyboard-driven view of agents, tasks, alerts, and activity
concord who # which agents are present and what they are working on
concord tasks # list all tracked tasks
concord handoff <task-id> # print the latest handoff
concord review-packet <id> # print the latest review packet
concord export markdown # regenerate .concord/ artifacts
concord doctor # workspace checks + per-task tool adoption
concord adapters status # global harness delivery capability matrix
concord --repo ../project status # select by repository path from anywhere
concord --workspace ws_... status # select an id returned by a Concord operation--repo y --workspace son opciones globales mutuamente excluyentes. La CLI usa la misma
prioridad CONCORD_REPO_ROOT → CLAUDE_PROJECT_DIR → directorio de trabajo
y la misma canonicalización de worktrees vinculados que MCP.
concord dashboard es una TUI local de pantalla completa de solo lectura. Se actualiza desde el
espacio de trabajo SQLite compartido cada segundo mientras mantiene agentes, tareas, alertas,
contexto y línea de tiempo dentro de un viewport de terminal fijo. Usa Tab para cambiar
de panel, j/k o las teclas de flecha para seleccionar trabajo, / para filtrar, ? para ayuda
y q para salir.
npm install -g @concord-ai/concord-mcp@latest
concord --versionConcord comprueba diariamente y muestra las actualizaciones disponibles en la CLI, las herramientas MCP y el
dashboard; concord setup puede instalar una con confirmación, y
CONCORD_NO_UPDATE_CHECK=1 desactiva las comprobaciones.
Estado de trabajo compartido y memoria de tareas para agentes de codificación que usan el mismo checkout local. No es un orquestador, revisor de código, servicio de sincronización alojado, base de datos vectorial de memoria ni un agente de codificación autónomo.
Ver también: ¿Por qué no usar simplemente markdown?
Consulta CONTRIBUTING.md y CLAUDE.md. Este
repositorio está estrictamente tipado (sin any, sin typecasts) y es modular. Las buenas primeras
tareas están etiquetadas como good first issue.
Concord envía telemetría de producto y coordinación a getconcord.ai. Incluye identificadores aleatorios de instalación/invocación; seudónimos irreversibles de espacio de trabajo y flujo de tareas por instalación; versiones de Concord/Node/plataforma; metadatos de cliente normalizados; nombres de operación, resultados y duraciones; resultados agregados de solapamiento/guardia de edición; etapas y latencias de entrega de mensajes; transiciones del ciclo de vida de tareas y tiempo transcurrido; y resultados de aceptación, integración, intervención humana y retrabajo notificados explícitamente.
Concord nunca envía código, rutas de archivo o repositorio sin procesar, remotos, nombres de usuario, identificadores de tarea o agente sin procesar, identificadores o contenido de mensajes, argumentos de comandos, entradas/salidas de herramientas ni contenido de tareas. El servidor receptor almacena la dirección IP de la solicitud y deriva/almacena un código de país. Esos campos del lado del servidor actualmente no tienen caducidad automática. Establece CONCORD_TELEMETRY_DISABLED=1 (o DO_NOT_TRACK=1) para desactivar la telemetría. La entrega es de mejor esfuerzo y nunca puede hacer fallar una operación de Concord.
Record completion evidence and atomically leave the task active as a handoff, mark it review-ready, or close it with a complete/closed outcome.
| Name | Required | Description | Default |
|---|---|---|---|
| reason | No | Audited terminal reason; defaults to the change summary | |
| outcome | No | Final task state; handoff records evidence without changing lifecycle state | complete |
| task_id | Yes | Stable task identifier, e.g. TASK-12 | |
| agent_id | No | Usually omit — Concord derives your identity from your session. Pass only the id your client told you (Codex); a session Concord can see always wins. | |
| decisions | No | Notable decisions and why | |
| diff_size | No | Rough diff size, e.g. +120 / -30 | |
| tests_run | No | Test commands run | |
| next_steps | No | Remaining work or follow-ups | |
| provenance | No | Evidence source for review claims | |
| assumptions | No | Assumptions made | |
| known_risks | No | Known risks introduced | |
| what_changed | Yes | Concise summary of what changed | |
| workspace_id | No | Workspace id returned by a Concord operation. Omit to use the automatically resolved repository workspace. | |
| changed_files | No | Files that changed | |
| open_questions | No | Unresolved review questions | |
| expected_version | Yes | Task version last read by the caller; stale versions are rejected | |
| reported_outcome | No | Optional measured acceptance, integration, or intervention result | |
| needs_review_from | No | Who should review | |
| guardrails_checked | No | Guardrails checked |
Does the description disclose side effects, auth requirements, rate limits, or destructive behavior?
No annotations are provided, so the description carries the full burden. It discloses atomicity and that the handoff outcome does not change lifecycle state, which is useful. However, it does not mention other behaviors like version staleness rejection, permission requirements, or idempotency—only the schema covers these.
Agents need to know what a tool does to the world before calling it. Descriptions should go beyond structured annotations to explain consequences.
Is the description appropriately sized, front-loaded, and free of redundancy?
The description is a single, well-structured sentence that front-loads the core action and concisely lists the three outcome alternatives. Every word earns its place without redundancy.
Shorter descriptions cost fewer tokens and are easier for agents to parse. Every sentence should earn its place.
Given the tool's complexity, does the description cover enough for an agent to succeed on first attempt?
Given the 19-parameter schema with full descriptions and no output schema, the description provides key behavioral context (recording evidence and state transitions). It is adequate for the tool's primary purpose, though it leaves the return value or side effects unexplained, which is a minor gap.
Complex tools with many parameters or behaviors need more documentation. Simple tools need less. This dimension scales expectations accordingly.
Does the description clarify parameter syntax, constraints, interactions, or defaults beyond what the schema provides?
Schema coverage is 100%, so the baseline is 3. The description adds minimal semantic value beyond the schema; it ties outcomes to lifecycle changes, but that is already documented in the outcome property. No additional parameter context is offered.
Input schemas describe structure but not intent. Descriptions should explain non-obvious parameter relationships and valid value ranges.
Does the description clearly state what the tool does and how it differs from similar tools?
The description states a specific verb ('record') and resource ('completion evidence') and enumerates the three possible task state outcomes. It clearly identifies the tool as the finalization step, though it does not explicitly contrast with update_work or transfer_work.
Agents choose between tools based on descriptions. A clear purpose with a specific verb and resource helps agents select the right tool.
Does the description explain when to use this tool, when not to, or what alternatives exist?
The description explains what the tool does but provides no explicit guidance on when to use it versus the sibling tools (e.g., update_work, transfer_work). The outcome choices are described, but the description lacks conditions or exclusions that would help an agent decide to pick this tool over alternatives.
Agents often have multiple tools that could apply. Explicit usage guidance like "use X instead of Y when Z" prevents misuse.
Read the workspace, one task, one agent communication inbox/outbox, or one durable prompt/reply thread by supplying at most one selector.
| Name | Required | Description | Default |
|---|---|---|---|
| task_id | No | Task to inspect; omit for the whole workspace state | |
| agent_id | No | Agent communication inbox/outbox to inspect | |
| message_id | No | Prompt/reply thread to inspect | |
| workspace_id | No | Workspace id returned by a Concord operation. Omit to use the automatically resolved repository workspace. |
Does the description disclose side effects, auth requirements, rate limits, or destructive behavior?
The annotations already declare this as read-only, idempotent, and non-destructive, so the description correctly aligns with them. Beyond that, it adds the constraint that only one selector can be supplied at a time and clarifies the different scopes (workspace, task, agent, message). This additional behavioral context is valuable and not present in the annotations.
Agents need to know what a tool does to the world before calling it. Descriptions should go beyond structured annotations to explain consequences.
Is the description appropriately sized, front-loaded, and free of redundancy?
The description is a single, well-structured sentence that front-loads the core action ('Read') and lists the scope. There is zero fluff; every word earns its place, and the selector constraint is clearly stated at the end.
Shorter descriptions cost fewer tokens and are easier for agents to parse. Every sentence should earn its place.
Given the tool's complexity, does the description cover enough for an agent to succeed on first attempt?
For a read-only inspection tool with four optional parameters, the description adequately explains the various selection modes and the workspace resolution behavior. It does not describe the return format, but given the tool name and the read-only nature, the output is self-evident. With no output schema, this would be a minor gap, but the description covers enough for correct invocation.
Complex tools with many parameters or behaviors need more documentation. Simple tools need less. This dimension scales expectations accordingly.
Does the description clarify parameter syntax, constraints, interactions, or defaults beyond what the schema provides?
Schema description coverage is 100%, so each parameter already has a clear description. The tool description adds the global constraint of 'at most one selector' and restates the resource types, but it does not provide deeper semantics beyond the schema. This meets the baseline for a well-covered schema.
Input schemas describe structure but not intent. Descriptions should explain non-obvious parameter relationships and valid value ranges.
Does the description clearly state what the tool does and how it differs from similar tools?
The description uses a specific verb 'Read' and lists the distinct resources (workspace, task, agent inbox/outbox, prompt/reply thread), making the tool's purpose unmistakable. It also clearly distinguishes it from the sibling tools (start_work, update_work, etc.) which are all actions, while this is the only inspection tool.
Agents choose between tools based on descriptions. A clear purpose with a specific verb and resource helps agents select the right tool.
Does the description explain when to use this tool, when not to, or what alternatives exist?
The description implies usage by stating that at most one selector can be supplied, which is useful for invoking the tool correctly. However, it does not explicitly state when to use this tool versus alternatives, nor does it provide any exclusions or conditions. The distinction from siblings is obvious from the tool names, but the description itself does not verbalize it.
Agents often have multiple tools that could apply. Explicit usage guidance like "use X instead of Y when Z" prevents misuse.
Enter one task before editing: register or refresh this agent, accept addressed work when needed, claim the declared scope, and return actionable overlap warnings.
| Name | Required | Description | Default |
|---|---|---|---|
| cwd | No | Agent working directory | |
| pid | No | Agent process id, if known | |
| kind | Yes | Agent type or provider, e.g. claude-code or codex | |
| model | No | Model the agent is running | |
| notes | No | Concise task notes | |
| owner | No | Human accountable for this agent and task | |
| title | Yes | Short human-readable title | |
| branch | No | Git branch, if known | |
| domains | No | Product domains touched | |
| modules | No | Logical modules touched | |
| summary | No | One-line description of the current work | |
| task_id | Yes | Stable task identifier, e.g. TASK-12 | |
| agent_id | No | Usually omit — Concord derives your identity from your session. Pass only the id your client told you (Codex); a session Concord can see always wins. | |
| worktree | No | Git worktree path, if used | |
| risk_tags | No | Risk tags shared with related work | |
| workspace_id | No | Workspace id returned by a Concord operation. Omit to use the automatically resolved repository workspace. | |
| expected_files | No | Files expected to change | |
| parent_task_id | No | Parent task for a smaller claimed unit |
Does the description disclose side effects, auth requirements, rate limits, or destructive behavior?
With no annotations provided, the description carries the full burden, and it does well: it discloses the multi-step side effects (register/refresh agent, accept addressed work, claim scope) and the return value ('actionable overlap warnings'). It communicates stateful behavior agents would not otherwise know from the schema.
Agents need to know what a tool does to the world before calling it. Descriptions should go beyond structured annotations to explain consequences.
Is the description appropriately sized, front-loaded, and free of redundancy?
A single compact sentence, efficiently front-loaded with the when ('Enter one task before editing') before the action list. The list of steps is dense but each item earns its place; no filler or repetition.
Shorter descriptions cost fewer tokens and are easier for agents to parse. Every sentence should earn its place.
Given the tool's complexity, does the description cover enough for an agent to succeed on first attempt?
For a complex 18-parameter workflow tool, the description captures the essential journey (register, accept, claim, return warnings) and names the output. The required parameters are covered by the schema, so nothing critical is missing for an agent to call it correctly.
Complex tools with many parameters or behaviors need more documentation. Simple tools need less. This dimension scales expectations accordingly.
Does the description clarify parameter syntax, constraints, interactions, or defaults beyond what the schema provides?
Schema coverage is 100%, so every parameter is already documented and the baseline is 3. The description's mention of 'claim the declared scope' and 'overlap warnings' loosely maps to workspace_id/risk_tags/expected_files, but it adds no concrete format or syntax guidance beyond what the schema provides.
Input schemas describe structure but not intent. Descriptions should explain non-obvious parameter relationships and valid value ranges.
Does the description clearly state what the tool does and how it differs from similar tools?
The description states a specific action ('Enter one task before editing: register or refresh this agent, accept addressed work when needed, claim the declared scope') with a clear resource and purpose. It implicitly differentiates from siblings (inspect/update/transfer/finish) by being the task-entry and scope-claiming action, though it doesn't name them explicitly.
Agents choose between tools based on descriptions. A clear purpose with a specific verb and resource helps agents select the right tool.
Does the description explain when to use this tool, when not to, or what alternatives exist?
The phrase 'Enter one task before editing' establishes when to use the tool as a pre-edit registration step, and 'accept addressed work when needed' gives conditional context. However, it never names alternatives such as update_work or finish_work or states when NOT to use it, leaving selection to inference.
Agents often have multiple tools that could apply. Explicit usage guidance like "use X instead of Y when Z" prevents misuse.
Apply one versioned ownership action: assign, accept, decline, release, reassign, offer an evidence-bearing handoff, or reopen terminal work.
| Name | Required | Description | Default |
|---|---|---|---|
| force | No | ||
| action | Yes | Ownership action to apply through the versioned task state machine | |
| reason | No | ||
| task_id | Yes | Stable task identifier, e.g. TASK-12 | |
| agent_id | No | Usually omit — Concord derives your identity from your session. Pass only the id your client told you (Codex); a session Concord can see always wins. | |
| decisions | No | Notable decisions and why | |
| tests_run | No | Test commands run | |
| handoff_id | No | Pending handoff to resolve; inferred from the task when omitted | |
| next_steps | No | Remaining work or follow-ups | |
| assumptions | No | Assumptions made | |
| known_risks | No | Known risks introduced | |
| to_agent_id | No | Required for assign, reassign, and offer | |
| what_changed | No | Required for offer | |
| workspace_id | No | Workspace id returned by a Concord operation. Omit to use the automatically resolved repository workspace. | |
| changed_files | No | Files that changed | |
| lease_seconds | No | ||
| expires_seconds | No | ||
| expected_version | Yes | Task version last read by the caller; stale versions are rejected | |
| guardrails_checked | No | Guardrails checked |
Does the description disclose side effects, auth requirements, rate limits, or destructive behavior?
With no annotations provided, the description carries the full burden of behavioral disclosure. It hints at 'versioned' and 'evidence-bearing handoff' but does not explain what versioning entails, whether actions are reversible, what state transitions occur, or what happens on failure. The description is too sparse to convey the operational impact of this mutation tool.
Agents need to know what a tool does to the world before calling it. Descriptions should go beyond structured annotations to explain consequences.
Is the description appropriately sized, front-loaded, and free of redundancy?
The description is a single, front-loaded sentence that efficiently conveys the core action set and concept. It contains no filler and gets to the point immediately. For a tool with 19 parameters, it is appropriately terse, though it could arguably benefit from more structure to improve scannability.
Shorter descriptions cost fewer tokens and are easier for agents to parse. Every sentence should earn its place.
Given the tool's complexity, does the description cover enough for an agent to succeed on first attempt?
Given the tool's complexity (19 parameters, no output schema, no annotations), the description is severely incomplete. It does not explain the versioning mechanism, the required fields for each action, the nature of an 'evidence-bearing handoff', or how these actions fit into the broader workflow with siblings. An agent would need to infer too much to use it correctly and safely.
Complex tools with many parameters or behaviors need more documentation. Simple tools need less. This dimension scales expectations accordingly.
Does the description clarify parameter syntax, constraints, interactions, or defaults beyond what the schema provides?
Schema description coverage is 79%, which is high, so the baseline is 3. The tool description adds no parameter-specific details beyond listing the action enum values, which are already present in the schema. Since the schema already documents parameters like task_id, action, expected_version, and to_agent_id, the description's lack of parameter elaboration does not significantly hinder understanding.
Input schemas describe structure but not intent. Descriptions should explain non-obvious parameter relationships and valid value ranges.
Does the description clearly state what the tool does and how it differs from similar tools?
The description states a clear verb ('Apply') and a specific resource ('versioned ownership action') and enumerates all seven allowed actions, making the tool's purpose unambiguous. It does not explicitly contrast with sibling tools, but the action list clearly distinguishes it from start_work, update_work, and finish_work, so the purpose is well-defined.
Agents choose between tools based on descriptions. A clear purpose with a specific verb and resource helps agents select the right tool.
Does the description explain when to use this tool, when not to, or what alternatives exist?
There is no guidance on when to use this tool versus its siblings. The description does not mention that this is for transferring ownership between agents, nor does it reference start_work, update_work, or finish_work, or any conditions that would make this tool the right choice. An agent is left to infer usage context.
Agents often have multiple tools that could apply. Explicit usage guidance like "use X instead of Y when Z" prevents misuse.
Record task context or deliver a live prompt/reply to another promptable workspace agent.
| Name | Required | Description | Default |
|---|---|---|---|
| kind | No | Kind of task-scoped update | |
| content | Yes | Concise context another agent needs | |
| task_id | No | Required for record; optional context for prompts | |
| agent_id | No | Usually omit — Concord derives your identity from your session. Pass only the id your client told you (Codex); a session Concord can see always wins. | |
| operation | No | Defaults to record; prompt and reply deliver live inter-agent messages | |
| to_agent_id | No | Recipient required for prompt | |
| workspace_id | No | Workspace id returned by a Concord operation. Omit to use the automatically resolved repository workspace. | |
| delivery_mode | No | Live delivery mode; steer is the only mode in v1 | |
| idempotency_key | No | Required for prompt and reply; makes delivery safe to retry | |
| reply_to_message_id | No | Message being answered for reply |
Does the description disclose side effects, auth requirements, rate limits, or destructive behavior?
With no annotations provided, the description carries the full burden of behavioral disclosure. It mentions 'live' delivery for prompts/replies but does not disclose idempotency requirements, delivery semantics, persistence of records, or any side effects. The schema covers idempotency_key requirement for prompt/reply, but that is in the schema, not this description.
Agents need to know what a tool does to the world before calling it. Descriptions should go beyond structured annotations to explain consequences.
Is the description appropriately sized, front-loaded, and free of redundancy?
The description is very short and front-loaded, which is efficient, but for a tool with 10 parameters and 3 operations it is too underspecified to be adequately informative. It is not verbose, but it fails to convey the operational complexity or usage nuances, so it earns a middle score.
Shorter descriptions cost fewer tokens and are easier for agents to parse. Every sentence should earn its place.
Given the tool's complexity, does the description cover enough for an agent to succeed on first attempt?
This is a complex tool with 10 parameters, three distinct operations, no annotations, and no output schema. The description merely lists two high-level capabilities and provides no clarity on operation selection, delivery behavior, or any edge cases. It is grossly incomplete for an agent to use it correctly.
Complex tools with many parameters or behaviors need more documentation. Simple tools need less. This dimension scales expectations accordingly.
Does the description clarify parameter syntax, constraints, interactions, or defaults beyond what the schema provides?
Schema coverage is 100%, so the description adds no parameter-level detail beyond what the schema already provides. It does not clarify the distinction between operations (record/prompt/reply) or explain any parameter interactions, so the baseline of 3 is appropriate.
Input schemas describe structure but not intent. Descriptions should explain non-obvious parameter relationships and valid value ranges.
Does the description clearly state what the tool does and how it differs from similar tools?
The description states a specific verb ('Record' and 'deliver') with a resource ('task context' and 'live prompt/reply'), clearly distinguishing this tool from siblings like start_work and finish_work by its messaging capability. However, it doesn't explicitly name sibling alternatives, so it stops short of full differentiation.
Agents choose between tools based on descriptions. A clear purpose with a specific verb and resource helps agents select the right tool.
Does the description explain when to use this tool, when not to, or what alternatives exist?
No guidance is given on when to use this tool versus alternatives; the description only states what it does without any conditions, prerequisites, or exclusions. It does not explain which operation (record vs. prompt vs. reply) to choose based on context.
Agents often have multiple tools that could apply. Explicit usage guidance like "use X instead of Y when Z" prevents misuse.
Recent tool additions, removals, and schema changes observed during successful MCP inspections. Dates show when Glama detected each change.
v0.1.0finish_workinspect_workstart_worktransfer_workupdate_workEach tool targets a distinct phase of the work lifecycle: starting, inspecting, updating, transferring ownership, and finishing. The actions are clearly separated with no overlap or ambiguous boundaries.
All tool names follow a consistent verb_noun pattern with 'work' as the noun, using only snake_case and imperative verbs (start, inspect, update, transfer, finish). The pattern is fully uniform.
Five tools is well-scoped for a workflow management server. Each tool covers a necessary stage without redundancy or bloat, making the set easy to navigate.
The tool set covers the entire task lifecycle: start, inspect (read), update (modify/communicate), transfer (ownership changes), and finish (close/handoff). It handles all major operations including reopening via transfer_work, so no critical gaps are apparent.
Real-time chat for AI agents. Claude Code, Cursor, Cline and Codex join channels over MCP.
Real-time chat hub for AI agents — Claude Code, Cursor, Cline, Codex over MCP or REST.
One MCP endpoint for Claude, GPT & Gemini: 100+ tools + no-code connectors + agent workers.
One identity across Claude Code, Codex, Cursor, Gemini, Windsurf: shared inbox and handoffs.
We provide all the information about MCP servers via our MCP API.
curl -X GET 'https://glama.ai/api/mcp/v1/servers/Get-Concord-AI/concord-mcp'
If you have feedback or need assistance with the MCP directory API, please join our Discord server
