Un servidor MCP que sirve la especificación canónica de Tank Fight: tamaño del campo y de la ventana, estadísticas de tanques y balas, reglas de aparición, puntuación, controles, comportamiento de los menús, estrategias de IA, la rampa de dificultad y un laberinto destruible por nivel.
Se ejecuta como un servicio local en un puerto. Cualquier cliente MCP puede conectarse, preguntar qué herramientas ofrece y consultar la especificación — mientras se construye el juego en Java, en TypeScript o en cualquier otro lenguaje.
Dos tareas, y la segunda importa más de lo que parece a primera vista.
1. Mantener las implementaciones independientes de acuerdo. Dos compilaciones del mismo juego, escritas por separado, divergen en el momento en que necesitan un número que nadie anotó. Que ambas pregunten al mismo servicio qué tamaño tiene un tanque elimina las conjeturas.
2. Evitar que el modelo invente requisitos. Cada detalle que una especificación omite es un lugar donde un asistente de codificación de IA elige silenciosamente algo plausible y sigue adelante — ¿las balas se detienen en las paredes? ¿pueden los aliados dispararse entre sí? ¿qué ocurre cuando un jugador es destruido a mitad de ronda? Rara vez menciona que lo decidió. Cada una de esas preguntas se responde aquí, de modo que la cuestión se convierte en una consulta en lugar de una suposición, y open_questions nombra lo que genuinamente no está decidido, para que la respuesta honesta sea "pregunta a un humano" en lugar de una invención con total seguridad.
Related MCP server: 17-0 MCP Server
Requiere uv.
uv syncuv run server.pyEso sirve la especificación en http://127.0.0.1:8082/mcp. Usa --port para escuchar en otro puerto, --host para enlazar otra interfaz y --stdio para clientes que solo hablan stdio.
Cualquier cliente MCP, cualquier número de ellos a la vez. Para Claude Code:
claude mcp add --transport http tank-spec http://localhost:8082/mcpAñade --scope project para registrarlo en el repositorio en el que estás trabajando, de modo que todos los que trabajen en esa implementación obtengan la misma especificación. Luego /mcp para confirmar la conexión, y el cliente puede preguntar al servidor qué herramientas existen y cómo llamarlas — aquí no hay nada que configurar por cliente.
Herramienta | Respuestas |
| Qué hay en la especificación, una línea por cada sección — empieza aquí |
| Una sección, o una ruta con puntos como |
| "¿Ocurre X?" — encuentra la regla que lo decide |
| Qué significa el nivel N: enemigos totales, máximo simultáneo, intervalo de aparición, qué laberinto |
| La disposición de ladrillos de ese nivel, expandida en casillas, con una imagen ASCII |
| Lo que la especificación deliberadamente no ha decidido |
No tiene idea de quién está conectado ni de qué están construyendo, y eso es a propósito. Conoce la especificación; los clientes conocen el servidor. Nada apunta en la otra dirección.
Así que comprobar si una implementación sigue coincidiendo con la especificación es tarea del cliente. Un cliente puede leer los valores de su propio código y compararlos con lo que devuelven estas herramientas — está en ese repositorio y el servidor no. Un servidor que se metiera en un checkout para inspeccionarlo tendría que conocer el lenguaje, la estructura de archivos y la ruta, y necesitaría actualizarse cada vez que apareciera otra implementación.
El mismo razonamiento es por el que la especificación se sirve en lugar de distribuirse como archivo:
Está fuera de cada implementación. Una sesión que trabaja en un repositorio no puede leer los archivos de otro, pero puede llamar a un servicio.
Se consulta, no se vuelca. search_spec("friendly fire") devuelve una regla. Cargar toda la especificación en el contexto de un modelo para responder una pregunta es exactamente el desperdicio que esto evita.
Algunas respuestas se calculan, no se almacenan. Los niveles se generan a partir de una fórmula de rampa — el nivel 6 no existe en ningún archivo. resolve_level es la única forma correcta de preguntar, y su truncamiento de enteros es esencial: una versión en coma flotante discrepa en cuatro de los ocho niveles.
Todo lo que está dentro del borde es ladrillo, y el ladrillo puede ser destruido a tiros — por ambos bandos, ya que los bandos importan para los tanques y nunca para las paredes. Una casilla es de 20x20 y recibe 40 de daño, así que dos balas estándar abren un agujero por el que luego pasan los tanques. El borde en sí es inmune.
Cada nivel tiene su propio laberinto, así que el número de nivel es la única perilla de dificultad: determina cuántos enemigos llegan, con qué rapidez y el terreno en el que aparecen. El nivel 1 es casi abierto; el nivel 8 es lo bastante denso como para que abrir un camino a tiros suela ser más rápido que encontrarlo.
get_maze(level) devuelve los rectángulos, la lista de casillas expandida y una imagen:
########################################
#EE................EE................EE#
#EE................EE................EE#
#......................................#
#......BBBBBBBBBBB....BBBBBBBBBBB......#
...
#.............AA...AA...AA.............#
########################################Usa las casillas expandidas en lugar de expandir tú mismo los rectángulos — ese paso, y la regla de qué casilla daña una bala cuando está a caballo entre dos, son donde dos compilaciones terminan más fácilmente cavando agujeros distintos con los mismos disparos.
Añadir un laberinto es solo añadir rectángulos bajo mazes.levels en el YAML. El conjunto de pruebas lo somete entonces a los invariantes que hacen que un nivel sea jugable: alineado con las casillas, dentro del área, sin solapamientos, cajas de aparición y de aliados despejadas, y todos los puntos de entrada accesibles a pie sin destruir nada — cavar es un atajo, nunca un requisito.
spec/game-spec.yaml es todo el contenido, y está pensado para editarse a mano — ese es el punto. El servidor lo relee en cada llamada, así que un cambio surte efecto de inmediato sin reiniciar nada.
La especificación se sembró a partir de una implementación funcional y es independiente de ella desde ese momento. Es la que manda: cambia un valor aquí primero y luego haz que las implementaciones lo sigan.
uv run pytesttests/test_spec.py fija los valores de la especificación y somete cada laberinto a los invariantes de jugabilidad. tests/test_server.py maneja el servidor a través de un cliente MCP — descubre las herramientas, lee sus descripciones, llámalas. tests/test_http.py inicia el proceso real en un puerto y se conecta a él por HTTP, como hace un cliente.
spec/game-spec.yaml the specification — the actual source of truth
server.py MCP server: tool definitions, their descriptions, and the transport
tank_spec/spec.py loading, searching, level maths, maze expansion
tests/ spec and maze invariants, MCP integration, HTTP transportGet a level's brick layout: the rectangles, and those rectangles expanded into the exact 20x20 tiles the level starts with.
Every level has its own maze, and all of the brick in it is destructible — bullets from either side chew through it, so the terrain changes as a round is played. Border walls are separate and indestructible.
Use the expanded tile list rather than expanding the rectangles yourself: that expansion is the step two implementations most easily get subtly different, and a one-tile disagreement changes which shots open a route.
Args: level: The level number. Out-of-range values are clamped. ascii_map: Include a picture of the layout — '#' border, 'B' brick, 'E' enemy entry point, 'A' ally start. Useful for checking a build renders the level you think it does; set false to save context.
| Name | Required | Description | Default |
|---|---|---|---|
| level | Yes | ||
| ascii_map | No |
| Name | Required | Description |
|---|---|---|
| result | Yes |
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 that all bricks are destructible and terrain changes during play, and that border walls are indestructible. It also notes that out-of-range level values are clamped, which is a behavioral trait. However, it doesn't mention the return format or any potential side effects, but given the tool is a read operation, this is adequate.
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 well-structured with a clear opening, a rationale for using the expanded tiles, and a concise Args section. Every sentence adds value, and it's appropriately sized for the tool's complexity.
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 has an output schema (not shown but mentioned), the description doesn't need to explain return values. It covers the key aspects: what the tool returns, why to use it, parameter semantics, and behavioral notes. The description is complete for an agent to select and invoke the tool 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 description coverage is 0%, so the description must compensate. It explains the 'level' parameter (level number, out-of-range clamped) and 'ascii_map' (include a picture, useful for checking build, set false to save context). This adds meaning beyond the schema's basic types and defaults.
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 clearly states the tool's purpose: to get a level's brick layout as rectangles and expanded 20x20 tiles. It specifies the resource (level maze) and the action (get), and distinguishes it from siblings by focusing on the brick layout, while siblings like list_sections and get_spec handle other aspects.
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 explicitly advises using the expanded tile list rather than expanding rectangles manually, explaining the risk of subtle differences. It also provides guidance on when to set ascii_map to false to save context, and mentions that border walls are separate and indestructible, which helps the agent understand the tool's output.
Agents often have multiple tools that could apply. Explicit usage guidance like "use X instead of Y when Z" prevents misuse.
Return one section of the Tank Fight specification as JSON.
Use this to get the authoritative numbers and rules before implementing a feature — screen size, tank and bullet stats, wall layout, controls, AI parameters, and so on. Do not infer these values from an existing implementation and do not invent them; this file is what the implementations are supposed to agree with.
Args: section: A section name from list_sections, or a dotted path to go straight to a value, e.g. "entities.bullet" or "ai.enemy.turn_chance_per_tick".
| Name | Required | Description | Default |
|---|---|---|---|
| section | Yes |
| Name | Required | Description |
|---|---|---|
| result | Yes |
Does the description disclose side effects, auth requirements, rate limits, or destructive behavior?
With no annotations, the description carries the full burden of behavioral disclosure. It adds crucial context: the spec is authoritative, and it explicitly warns against inferring values from existing implementations. However, it does not mention error handling or behavior for invalid section names, so it's not fully transparent on edge cases.
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 compact and front-loaded, with a clear one-sentence summary followed by usage guidance and a well-formatted Args section. Every sentence adds value, and the formatting makes it easy to scan.
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 simple single-parameter tool with an output schema, the description is complete: it explains the purpose, parameter format, and authoritative nature of the data. It appropriately references list_sections for valid section names and provides examples of dotted paths. No additional return-value details are needed given the output schema.
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?
The input schema only declares 'section' as a string with no description, so the description provides all parameter semantics. It explains that the value can be a section name from list_sections or a dotted path, with concrete examples like 'entities.bullet'. This is essential information not available from the schema alone.
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 opens with 'Return one section of the Tank Fight specification as JSON,' which is a specific verb+resource statement. It also explains the use case ('before implementing a feature') and distinguishes itself from sibling tools by focusing on retrieving a named section of the specification.
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?
It clearly states when to use the tool ('before implementing a feature') and warns against inferring or inventing values, which provides strong context. However, it does not explicitly name alternative tools or exclusions beyond referencing list_sections for section names, so it stops short of the explicit alternative guidance found in a 5.
Agents often have multiple tools that could apply. Explicit usage guidance like "use X instead of Y when Z" prevents misuse.
List the sections of the Tank Fight specification, with a one-line summary of each, so you can fetch only the part you need.
Start here when you don't yet know where a rule lives. Fetching the whole spec is rarely necessary and wastes context.
| Name | Required | Description | Default |
|---|---|---|---|
No parameters | |||
| Name | Required | Description |
|---|---|---|
| result | Yes |
Does the description disclose side effects, auth requirements, rate limits, or destructive behavior?
With no annotations provided, the description carries the transparency burden. It clearly discloses that the tool returns a section list with summaries rather than full content, and frames itself as a navigation helper. For a simple read-only listing tool, this is sufficient behavioral context.
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 two sentences long, front-loaded with the core purpose, and every phrase earns its place. It avoids filler while adding practical usage context.
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 zero-parameter listing tool with an output schema, the description is complete: it says what the tool returns, why to use it, and when to start with it. It sufficiently covers the low-complexity context.
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?
The tool has zero parameters and schema coverage is effectively 100%, so no parameter explanation is needed. The baseline of 4 applies because there is no param information to compensate for.
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 clearly identifies the action ('List the sections'), the resource ('Tank Fight specification'), and the output form ('with a one-line summary of each'). It also differentiates from siblings by positioning this as the starting point for locating rules before fetching a specific part.
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?
It gives explicit guidance: 'Start here when you don't yet know where a rule lives' and warns that 'Fetching the whole spec is rarely necessary and wastes context.' It lacks an explicit contrast with search_spec, which is a sibling that may be a better alternative when users know search terms.
Agents often have multiple tools that could apply. Explicit usage guidance like "use X instead of Y when Z" prevents misuse.
List the design decisions the specification deliberately has NOT made.
Check this before inventing an answer to something the spec is silent on. If your question is here, the honest answer is "not decided yet" — say so and ask, rather than picking something and moving on. Anything on this list is a place where two implementations will otherwise diverge.
| Name | Required | Description | Default |
|---|---|---|---|
No parameters | |||
| Name | Required | Description |
|---|---|---|
| result | Yes |
Does the description disclose side effects, auth requirements, rate limits, or destructive behavior?
The description adds behavioral context: the tool returns a list of open items and hints that these are places where implementations diverge. However, no annotations are provided, so the description carries the burden. It does not state whether the list is static or generated, or any caching/rate limits, but for this kind of read-only tool, it is adequate. No contradiction.
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?
Four short paragraphs, each earning its place: purpose, timing, action on hit, consequence. No fluff, front-loaded with the core action.
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 simple zero-parameter tool, the description fully explains when and how to use it and what it returns. It doesn't describe output format, but an output schema exists, so that's not needed. It is complete for its complexity.
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?
The tool has zero parameters and schema coverage is 100%, so the description has little to add. The description explains what the output represents (open design decisions), adding semantic meaning beyond the empty schema. Baseline for 0 params is 4, and it meets that.
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 clearly states the tool lists design decisions left open by the specification. It uses specific verb 'List' and resource 'design decisions', and distinguishes it from sibling tools like list_sections or get_spec by focusing on unresolved questions.
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?
Provides explicit guidance: check this tool before inventing an answer, and what to do if a question is here (say 'not decided yet' and ask). It implies when not to use (when question is not listed, it may be intentionally handled elsewhere).
Agents often have multiple tools that could apply. Explicit usage guidance like "use X instead of Y when Z" prevents misuse.
Resolve what a level number actually means: how many enemy tanks the round fields in total, how many may be alive at once, and how often they arrive.
Levels are GENERATED from a ramp formula, not stored anywhere, so this cannot be answered by reading a file — call this tool instead of computing it yourself, because the formula truncates with integer division and a floating-point version disagrees at several levels.
The result also names the level's maze; call get_maze for its actual layout.
Args: level: The level number. Out-of-range values are clamped to the supported range rather than rejected.
| Name | Required | Description | Default |
|---|---|---|---|
| level | Yes |
| Name | Required | Description |
|---|---|---|
| result | Yes |
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 reveals important traits: integer division with truncation, potential disagreement with floating-point calculations, and clamping behavior for out-of-range values. It also states what the result includes without exposing unnecessary details. This goes beyond a generic description and provides genuinely useful edge-case information.
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 appropriately concise, with a clear lead sentence, a paragraph explaining rationale, and a compact Args section. Every sentence earns its place—no filler or redundancy. The structure (purpose, why, related tool, parameter details) is logical and easy to scan.
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 simplicity (one parameter) and the presence of an output schema, the description covers all necessary aspects: what it computes, why it's needed, how parameters behave, and what the output references (maze). It does not omit critical context and does not need to detail return types because an output schema is provided.
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?
The schema has no description for the 'level' parameter (0% coverage), so the description must compensate. The Args section does this effectively by explaining the parameter's meaning and its clamping behavior, which is not present in the schema. This fully addresses potential confusion and guides correct usage.
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 clearly states the tool's purpose with a specific verb ('Resolve') and a detailed resource ('what a level number actually means: how many enemy tanks the round fields in total, how many may be alive at once, and how often they arrive'). It distinguishes itself from siblings by mentioning get_maze for maze layout and contrasting with reading a file. This is more than just a restatement of the name—it provides concrete capabilities.
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 explicitly tells when to use this tool instead of alternatives: 'Levels are GENERATED from a ramp formula, not stored anywhere, so this cannot be answered by reading a file — call this tool instead of computing it yourself'. It also directs to a sibling ('call get_maze for its actual layout'). This gives clear decision guidance, exceeding basic when-to-use.
Agents often have multiple tools that could apply. Explicit usage guidance like "use X instead of Y when Z" prevents misuse.
Search the specification for the rule or value that answers a question.
Use this whenever you are about to assume how the game behaves — "do bullets stop at walls?", "can allies shoot each other?", "how fast is a bullet?", "what happens when a player is destroyed?". Searching costs far less context than loading whole sections, and the answer is authoritative.
If the search comes back empty, the spec may genuinely not cover it: check open_questions() before deciding anything yourself.
Args: query: Plain words describing what you need, e.g. "friendly fire" or "spawn interval" or "health bar colour".
| Name | Required | Description | Default |
|---|---|---|---|
| query | Yes |
| Name | Required | Description |
|---|---|---|
| result | Yes |
Does the description disclose side effects, auth requirements, rate limits, or destructive behavior?
No annotations are provided, so the description carries the burden. It explains the tool is a read-only search ('costs far less context than loading whole sections') and the behavior on empty results (spec may not cover it, check open_questions). This adds transparency about the search's scope and fallback, though it doesn't detail return format or limits.
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 appropriately concise, using a short intro, a usage guideline paragraph, and a parameter explanation. Every sentence adds value, and it is front-loaded with the core purpose.
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?
The tool is simple (one parameter) and the description covers purpose, usage, fallback behavior, and parameter semantics. The output schema exists, so return values don't need explanation. It is complete for a search tool.
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 0%, so the description must compensate. It explains the 'query' parameter with examples: 'Plain words describing what you need, e.g. "friendly fire" or "spawn interval" or "health bar colour".' This adds meaningful guidance beyond the schema's bare string type.
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 clearly states the tool's purpose: 'Search the specification for the rule or value that answers a question.' It uses a specific verb ('search') and resource ('the specification'), and provides concrete examples of questions it answers, distinguishing it from sibling tools like list_sections or get_spec.
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 explicitly guides when to use this tool: 'Use this whenever you are about to assume how the game behaves'. It contrasts with alternatives by noting that searching costs less context than loading whole sections. It also provides a clear fallback: if the search comes back empty, check open_questions() before deciding anything yourself.
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.
v0.1.0get_mazeget_speclist_sectionsopen_questionsresolve_levelsearch_specScored across 6 tools
Each tool has a distinct purpose: listing sections, getting a section, searching the spec, resolving level semantics, retrieving maze layouts, and listing open questions. No overlaps or ambiguity.
All tool names follow a consistent verb_noun pattern (list_, get_, search_, resolve_, open_), making the API predictable and easy to navigate.
Six tools is well-scoped for a specification server, covering all necessary access patterns without unnecessary redundancy or bloat.
The tool set fully covers the domain: discovering, retrieving, searching, interpreting, and visualizing spec details, plus handling open questions. No obvious gaps.
Read-only MCP server for the OPERANT AI operating-agent calibration benchmark.
Read-only MCP server for the WebAssembly spec: instructions, types, sections, search, proposals.
Public, read-only MCP server for FarmNeural company facts, packages, and capabilities.
Agent-native MCP server over the public saagarpatel.dev corpus. Read-only, stateless.