Un servidor de Protocolo de Contexto de Modelo (MCP) permite a asistentes de IA como Claude usar la API de Búsqueda de IA de Exa para búsquedas web. Esta configuración permite a los modelos de IA obtener información web en tiempo real de forma segura y controlada.
Vídeo de demostración https://www.loom.com/share/ac676f29664e4c6cb33a2f0a63772038?sid=0e72619f-5bfc-415d-a705-63d326373f60
El Protocolo de Contexto de Modelo (MCP) es un sistema que permite que las aplicaciones de IA, como Claude Desktop, se conecten a herramientas y fuentes de datos externas. Ofrece una forma clara y segura para que los asistentes de IA trabajen con servicios y API locales, manteniendo al usuario en control.
Related MCP server: Perplexity MCP Server
El servidor Exa MCP:
Permite que los asistentes de IA realicen búsquedas web utilizando la potente API de búsqueda de Exa
Proporciona resultados de búsqueda estructurados que incluyen títulos, URL y fragmentos de contenido.
Almacena en caché búsquedas recientes como recursos de referencia
Maneja con elegancia los casos de limitación de velocidad y de error
Admite el rastreo web en tiempo real para contenido nuevo
Antes de comenzar, asegúrese de tener:
Node.js (v18 o superior)
Claude Desktop instalado
Una clave API de Exa
Git instalado
Puede verificar su instalación de Node.js ejecutando:
node --version # Should show v18.0.0 or highernpm install -g exa-mcp-serverPara instalar el servidor Exa MCP para Claude Desktop automáticamente a través de Smithery :
npx -y @smithery/cli install exa --client claudeClonar el repositorio:
git clone https://github.com/exa-labs/exa-mcp-server.git
cd exa-mcp-serverInstalar dependencias:
npm installConstruir el proyecto:
npm run buildCrea un enlace global (esto hace que el servidor sea ejecutable desde cualquier lugar):
npm linkPuedes encontrar claude_desktop_config.json dentro de la configuración de la aplicación Claude Desktop:
Abra la aplicación Claude Desktop y habilite el Modo de desarrollador desde la barra de menú superior izquierda.
Una vez habilitado, abre Configuración (también desde la barra de menú superior izquierda) y navega hasta las Opciones de desarrollador, donde encontrarás el botón Editar configuración. Al hacer clic, se abrirá el archivo claude_desktop_config.json, que te permitirá realizar las modificaciones necesarias.
O (si desea abrir claude_desktop_config.json desde la terminal)
Abra la configuración de Claude Desktop:
code ~/Library/Application\ Support/Claude/claude_desktop_config.jsonAbra la configuración de Claude Desktop:
code %APPDATA%\Claude\claude_desktop_config.json{
"mcpServers": {
"exa": {
"command": "npx",
"args": ["/path/to/exa-mcp-server/build/index.js"],
"env": {
"EXA_API_KEY": "your-api-key-here"
}
}
}
}Reemplace your-api-key-here con su clave API de Exa real desde dashboard.exa.ai/api-keys .
Para que los cambios surtan efecto:
Salir completamente de Claude Desktop (no solo cerrar la ventana)
Inicie Claude Desktop nuevamente
Busque el ícono 🔌 para verificar que el servidor Exa esté conectado
Una vez configurado, puedes pedirle a Claude que realice búsquedas web. Aquí tienes algunos ejemplos:
Can you search for recent developments in quantum computing?Search for and summarize the latest news about artificial intelligence startups in new york.Find and analyze recent research papers about climate change solutions.Search for today's breaking news about tech.Search for the top 10 AI research papers from 2023, and only use live crawling as a fallback.Search for electric vehicles and return 3 results, always using live crawling.El servidor hará lo siguiente:
Procesar la solicitud de búsqueda
Consulta la API de Exa con configuraciones óptimas (incluido el rastreo en vivo)
Devolver resultados formateados a Claude
Almacenar en caché la búsqueda para futuras referencias
Herramienta de búsqueda web simplificada : permite a Claude buscar en la web con solo un parámetro de consulta
Parámetros de búsqueda personalizables : controle la cantidad de resultados y la estrategia de rastreo en vivo
Rastreo automático en vivo : utiliza rastreo en tiempo real según una estrategia específica
Parámetros óptimos preestablecidos : utiliza los mejores valores predeterminados para el recuento de resultados y los límites de caracteres
Almacenamiento en caché de búsqueda : guarda búsquedas recientes como recursos de referencia
Manejo de errores : maneja con elegancia los errores de API y los límites de velocidad
Seguridad de tipos : implementación completa de TypeScript con validación de Zod
Cumplimiento de MCP : implementa completamente la última especificación del protocolo MCP
Puede probar el servidor directamente utilizando el Inspector MCP:
npx @modelcontextprotocol/inspector node ./build/index.jsEsto abre una interfaz interactiva donde puede explorar las capacidades del servidor, ejecutar consultas de búsqueda y ver resultados de búsqueda almacenados en caché.
Servidor no encontrado
Verifique que el enlace npm esté configurado correctamente
Compruebe la sintaxis de configuración de Claude Desktop
Asegúrese de que Node.js esté instalado correctamente
Problemas con la clave API
Confirme que su EXA_API_KEY sea válida
Compruebe que EXA_API_KEY esté configurado correctamente en la configuración de Claude Desktop
Verifique que no haya espacios ni comillas alrededor de la clave API
Problemas de conexión
Reiniciar Claude Desktop por completo
Consultar los registros de Claude Desktop: GXP18
Si encuentra problemas, revise la documentación de MCP o visite las discusiones de GitHub para obtener soporte de la comunidad.
Exa AI por su potente API de búsqueda
Protocolo de contexto de modelo para la especificación MCP
Antrópico para Claude Desktop
Search and get relevant context for any programming task. Exa-code has the highest quality and freshest context for libraries, SDKs, and APIs. Use this tool for ANY question or task for related to programming. RULE: when the user's query contains exa-code or anything related to code, you MUST use this tool.
| Name | Required | Description | Default |
|---|---|---|---|
| query | Yes | Search query to find relevant context for APIs, Libraries, and SDKs. For example, 'React useState hook examples', 'Python pandas dataframe filtering', 'Express.js middleware', 'Next js partial prerendering configuration' | |
| tokensNum | No | Number of tokens to return (1000-50000). Default is 5000 tokens. Adjust this value based on how much context you need - use lower values for focused queries and higher values for comprehensive documentation. |
Does the description disclose side effects, auth requirements, rate limits, or destructive behavior?
Annotations already indicate read-only, idempotent, and non-destructive behavior. The description adds value by emphasizing 'highest quality and freshest context' and the programming domain focus, but doesn't disclose additional behavioral traits like rate limits, authentication needs, or response format details. No contradiction with annotations exists.
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 front-loaded with the core purpose and usage rule, but includes some redundancy (e.g., repeating 'exa-code' emphasis). Sentences are generally purposeful, though the 'RULE' phrasing could be more integrated. Overall efficient but with minor verbosity.
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 moderate complexity (2 parameters, no output schema), annotations cover safety aspects, and the description provides clear purpose and usage rules. However, it lacks details on response structure or error handling, which would enhance completeness 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 100%, providing full documentation for both parameters. The description doesn't add meaningful parameter semantics beyond what's in the schema, such as explaining query formulation strategies or token usage trade-offs. Baseline score of 3 is appropriate given the comprehensive 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 clearly states the tool's purpose: 'Search and get relevant context for any programming task' with specific focus on 'libraries, SDKs, and APIs.' It distinguishes from the sibling tool 'web_search_exa' by specifying programming-related content, though it doesn't explicitly contrast their differences beyond domain focus.
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 provides explicit usage guidance: 'Use this tool for ANY question or task related to programming' and includes a mandatory rule: 'when the user's query contains exa-code or anything related to code, you MUST use this tool.' This clearly defines when to use it versus alternatives, though it doesn't specify when NOT to use it for non-programming queries.
Agents often have multiple tools that could apply. Explicit usage guidance like "use X instead of Y when Z" prevents misuse.
Search the web using Exa AI - performs real-time web searches and can scrape content from specific URLs. Supports configurable result counts and returns the content from the most relevant websites.
| Name | Required | Description | Default |
|---|---|---|---|
| query | Yes | Websearch query | |
| numResults | No | Number of search results to return (default: 8) | |
| livecrawl | No | Live crawl mode - 'fallback': use live crawling as backup if cached content unavailable, 'preferred': prioritize live crawling (default: 'fallback') | |
| type | No | Search type - 'auto': balanced search (default), 'fast': quick results, 'deep': comprehensive search | |
| contextMaxCharacters | No | Maximum characters for context string optimized for LLMs (default: 10000) |
Does the description disclose side effects, auth requirements, rate limits, or destructive behavior?
Annotations already declare readOnlyHint=true, idempotentHint=true, and destructiveHint=false, covering safety and idempotency. The description adds valuable behavioral context beyond annotations: it mentions real-time web searches, scraping from specific URLs, configurable result counts, and returning content from relevant websites. This provides useful operational details without contradicting 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 appropriately sized at two sentences, front-loading the core purpose. Every sentence adds value: the first defines the tool's function, and the second elaborates on features and output. There's no wasted text, though it could be slightly more structured for optimal clarity.
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 moderate complexity, rich annotations (covering safety and idempotency), and 100% schema coverage, the description is reasonably complete. It explains the tool's function and key features. The lack of an output schema is a minor gap, but the description mentions return content, partially compensating. For a read-only search tool, this provides adequate 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?
Schema description coverage is 100%, so the schema fully documents all 5 parameters. The description adds minimal parameter semantics beyond the schema, mentioning only 'configurable result counts' (referencing numResults) and 'content from the most relevant websites' (hinting at query relevance). Since the schema does the heavy lifting, the baseline score 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 clearly states the tool's purpose: 'Search the web using Exa AI - performs real-time web searches and can scrape content from specific URLs.' It specifies the verb (search/scrape) and resource (web/URLs), making the function unambiguous. However, it doesn't explicitly differentiate from its sibling 'get_code_context_exa' beyond mentioning general web search 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 implies usage for web searches and content scraping, but provides no explicit guidance on when to use this tool versus its sibling 'get_code_context_exa' or other alternatives. It mentions configurable result counts and relevance, which suggests some context, but lacks clear when/when-not directives or named alternatives.
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.
v1.0.0company_research_exacrawling_exadeep_researcher_checkdeep_researcher_startget_code_context_exalinkedin_search_exaweb_search_exa5 fields changedInput schema / properties / contextMaxCharactersAdded value: +{
+ "description": "Maximum characters for context string optimized for LLMs (default: 10000)",
+ "type": "number"
+}Input schema / properties / livecrawlAdded value: +{
+ "description": "Live crawl mode - 'fallback': use live crawling as backup if cached content unavailable, 'preferred': prioritize live crawling (default: 'fallback')",
+ "enum": [
+ "fallback",
+ "preferred"
+ ],
+ "type": "string"
+}Input schema / properties / numResults / descriptionPrevious value: -"Number of search results to return (default: 5)"New value: +"Number of search results to return (default: 8)"Input schema / properties / query / descriptionPrevious value: -"Search query"New value: +"Websearch query"Input schema / properties / typeAdded value: +{
+ "description": "Search type - 'auto': balanced search (default), 'fast': quick results, 'deep': comprehensive search",
+ "enum": [
+ "auto",
+ "fast",
+ "deep"
+ ],
+ "type": "string"
+}company_research_exacrawling_exadeep_researcher_checkdeep_researcher_startlinkedin_search_exaweb_search_exaScored across 2 tools
The two tools have clearly distinct purposes: get_code_context_exa is specialized for programming-related searches with high-quality code context, while web_search_exa is a general web search tool for broader queries. There is no overlap in functionality, making it easy for an agent to choose the correct tool based on the query content.
Both tools follow a consistent naming pattern: they use snake_case and start with a verb (get, search) followed by a noun (code_context, web). The pattern is uniform across the set, with no deviations in style or structure.
With only 2 tools, the server feels thin for a general-purpose search domain, as it might lack coverage for intermediate or specialized tasks beyond code and web searches. However, the tools are well-defined, so it's borderline but not severely mismatched.
The server covers two key search domains (code and web), but there are notable gaps: it lacks tools for other common search types (e.g., image, news, academic) or advanced operations like filtering or saving results. This could limit agent effectiveness in broader search scenarios.
Connect AI agents to Exa for web search, content fetching, and multi-step research.
Provides AI assistants with access to Seltz's powerful Web Search capabilities.
Give AI assistants access to real-time data. Search the web, compare flights, find hotels, and more.
The best web search for your AI Agent
