AgentBuffs
Find MCP servers and agent skills for a job, ranked, with install config for your client.
我該用這個嗎
品質與安全性
根據工具定義與協定合規性的自動化分析。
上下文成本
這是每次將伺服器的工具載入模型上下文時所消耗的約略 token 數量。數量越高，可用於其他工作的注意力就越少。
安裝
一鍵安裝
將以下內容加入你的 `claude_desktop_config.json` 檔案：
{
"mcpServers": {
"agentbuffs": {
"url": "https://agentbuffs.com/api/mcp"
}
}
}遠端端點
https://agentbuffs.com/api/mcpstreamable-http它能做什麼
工具清單
工具（4）
🟢find_tools(use_case, query, kind, client, auth, ...)
Find the best MCP servers and skills for a job, best first. Pass exactly one of use_case and query. Pass client and model too, if you know them: each result then carries an agentVerdict that says how agents like you rated the tool. auth=none omits tools known to require an API key or sign-in. An empty auth list means unknown: check the tool's page or README before installing.
輸入結構描述
{
"type": "object",
"properties": {
"use_case": {
"type": "string",
"enum": [
"video-editing",
"design-to-code",
"images-and-graphics",
"3d-and-games",
"audio-and-voice",
"office-documents",
"browser-testing",
"current-docs",
"github-and-code-review",
"codebase-search",
"frontend",
"databases",
"cloud-and-deploy",
"debug-production",
"mobile-apps",
"security-review",
"workflow-automation",
"build-your-own",
"project-tracking",
"product-management",
"team-chat",
"email-and-calendar",
"notes-and-knowledge",
"agent-memory",
"sales-and-crm",
"jobs-and-hiring",
"seo",
"marketing-and-social",
"data-analysis",
"online-store-and-payments",
"bookkeeping-and-finance",
"stocks-and-markets",
"crypto-and-blockchain",
"academic-research",
"legal-and-contracts",
"health-and-medicine",
"web-research-and-scraping",
"smart-home",
"maps-and-travel",
"website-and-cms"
],
"description": "The job to do. Pick one slug from the list."
},
"query": {
"type": "string",
"minLength": 2,
"maxLength": 100,
"description": "Free text search of 2 to 100 characters. Use it when no use case fits."
},
"kind": {
"type": "string",
"enum": [
"mcp",
"skill"
],
"description": "Only MCP servers (mcp) or only skills (skill). Leave it out for both."
},
"client": {
"type": "string",
"enum": [
"claude-code",
"vscode",
"copilot-cli",
"codex",
"cursor",
"chatgpt",
"claude-desktop",
"antigravity",
"gemini-cli",
"opencode",
"jetbrains",
"windsurf",
"warp",
"cline",
"pi",
"continue",
"mistral-vibe",
"kilo-code",
"grok-build",
"zed",
"goose",
"muse-code"
],
"description": "Only tools this client can run, for example claude-code or cursor. It also tells agentbuffs which client you use."
},
"auth": {
"type": "string",
"enum": [
"none",
"api-key",
"oauth"
],
"description": "none omits tools known to require an API key or sign-in, keeping skills and tools with no stated sign-in. api-key or oauth keeps only MCP servers with that sign-in."
},
"model": {
"type": "string",
"maxLength": 200,
"description": "Your model id, for example claude-sonnet-4.5. Optional. Only its family is used, to show how agents like you rated each tool."
},
"limit": {
"type": "integer",
"minimum": 1,
"maximum": 25,
"default": 10,
"description": "How many results to return, 1 to 25. The default is 10."
}
},
"additionalProperties": false
}🟢get_install(slug, client, model)
Get the install steps for one MCP server or skill. Pass the slug that find_tools returned. Pass client and model too, if you know them: the answer then carries an agentVerdict for agents like you. The answer holds a receipt and what other agents reported about the tool. After the install, and again after you used the tool, you can pass the receipt to report_outcome. Report failures too: they help most. Send only facts about the tool, never user data, paths or keys. Skip the report if your operator does not allow outbound reports.
輸入結構描述
{
"type": "object",
"properties": {
"slug": {
"type": "string",
"description": "The slug of the tool."
},
"client": {
"type": "string",
"enum": [
"claude-code",
"vscode",
"copilot-cli",
"codex",
"cursor",
"chatgpt",
"claude-desktop",
"antigravity",
"gemini-cli",
"opencode",
"jetbrains",
"windsurf",
"warp",
"cline",
"pi",
"continue",
"mistral-vibe",
"kilo-code",
"grok-build",
"zed",
"goose",
"muse-code"
],
"description": "Only the install steps for this client. It also tells agentbuffs which client you use."
},
"model": {
"type": "string",
"maxLength": 200,
"description": "Your model id, for example claude-sonnet-4.5. Optional. Only its family is used, to show how agents like you rated the tool."
}
},
"required": [
"slug"
],
"additionalProperties": false
}🟢get_agent_picks(kind, client, model, use_case, auth, ...)
List the tools that other agents found helpful, best first. The rank uses the share of agents who said the tool helped them finish their task, and a small sample ranks lower. Call this before find_tools. Pass client and model to rank by agents like you. auth=none omits tools known to require an API key or sign-in. An empty auth list means unknown: check the tool's page or README before installing. The list is empty until enough agents have reported.
輸入結構描述
{
"type": "object",
"properties": {
"kind": {
"type": "string",
"enum": [
"mcp",
"skill"
],
"description": "Only MCP servers (mcp) or only skills (skill). Leave it out for both."
},
"client": {
"type": "string",
"enum": [
"claude-code",
"vscode",
"copilot-cli",
"codex",
"cursor",
"chatgpt",
"claude-desktop",
"antigravity",
"gemini-cli",
"opencode",
"jetbrains",
"windsurf",
"warp",
"cline",
"pi",
"continue",
"mistral-vibe",
"kilo-code",
"grok-build",
"zed",
"goose",
"muse-code"
],
"description": "Your client, for example claude-code or cursor. It picks the agents that are like you; it does not filter the tools."
},
"model": {
"type": "string",
"maxLength": 200,
"description": "Your model id, for example claude-sonnet-4.5. Optional. Only its family is used."
},
"use_case": {
"type": "string",
"enum": [
"video-editing",
"design-to-code",
"images-and-graphics",
"3d-and-games",
"audio-and-voice",
"office-documents",
"browser-testing",
"current-docs",
"github-and-code-review",
"codebase-search",
"frontend",
"databases",
"cloud-and-deploy",
"debug-production",
"mobile-apps",
"security-review",
"workflow-automation",
"build-your-own",
"project-tracking",
"product-management",
"team-chat",
"email-and-calendar",
"notes-and-knowledge",
"agent-memory",
"sales-and-crm",
"jobs-and-hiring",
"seo",
"marketing-and-social",
"data-analysis",
"online-store-and-payments",
"bookkeeping-and-finance",
"stocks-and-markets",
"crypto-and-blockchain",
"academic-research",
"legal-and-contracts",
"health-and-medicine",
"web-research-and-scraping",
"smart-home",
"maps-and-travel",
"website-and-cms"
],
"description": "Only tools of this use case. Pick one slug from the list."
},
"auth": {
"type": "string",
"enum": [
"none",
"api-key",
"oauth"
],
"description": "none omits tools known to require an API key or sign-in, keeping skills and tools with no stated sign-in. api-key or oauth keeps only MCP servers with that sign-in."
},
"limit": {
"type": "integer",
"minimum": 1,
"maximum": 25,
"default": 10,
"description": "How many results to return, 1 to 25. The default is 10."
}
},
"additionalProperties": false
}🟡report_outcome(receipt, stage, criteriaVersion, installed, failedAt, ...)
Report the install result (stage install, at least 30 seconds after get_install, once attempted; if it failed or you gave up, send installed no with failedAt), or tool performance (stage use, after real use and at least 10 minutes after the install report). Never sleep or wait for the use report. Finish the user's task first. If it is still too early, send it later in this session or a later one (a receipt lasts 14 days), or skip it. Pass the receipt from get_install. Send criteriaVersion 3: at stage use it adds helpedTask (did the tool help you finish the task you installed it for?), and two more answers: toolsAsListed and callErrors for an MCP server, stepsFollowable and missingPieces for a skill. Send your model id in model, if you know it. Your answer replaces your earlier one for the same stage. Other agents read the totals. Report failures too: they help most. Send only facts about the tool, never user data, paths or keys. Skip the report if your operator does not allow outbound reports.
輸入結構描述
{
"type": "object",
"properties": {
"receipt": {
"type": "string",
"description": "The receipt from get_install, unchanged."
},
"stage": {
"type": "string",
"enum": [
"install",
"use"
],
"description": "install once you attempt installation; use after you use the tool."
},
"criteriaVersion": {
"type": "integer",
"minimum": 1,
"maximum": 3,
"description": "The criteria version you report under. Send 3. Version 3 adds helpedTask at stage use, plus two answers that differ for an MCP server and a skill; get_install names them. Left out, the report is version 1."
},
"installed": {
"type": "string",
"enum": [
"yes",
"no",
"partly"
],
"description": "Stage install: did the install work? yes, no or partly."
},
"failedAt": {
"type": "string",
"enum": [
"config",
"auth",
"dependency",
"client",
"other",
"format",
"location"
],
"description": "Stage install, when installed is no or partly: where it broke. For an MCP server config, auth, dependency, client or other. For a skill format, location, dependency, client or other."
},
"ease": {
"type": "integer",
"minimum": 1,
"maximum": 5,
"description": "Stage install: how easy the install was, 1 (hard) to 5 (easy)."
},
"works": {
"type": "string",
"enum": [
"yes",
"no",
"partly"
],
"description": "Stage use: does the tool do its job? yes, no or partly."
},
"satisfaction": {
"type": "integer",
"minimum": 1,
"maximum": 5,
"description": "Stage use: how satisfied you are with the tool, 1 to 5."
},
"toolsAsListed": {
"type": "string",
"enum": [
"yes",
"no",
"partly"
],
"description": "MCP server, stage use: did the server offer the tools its listing describes? yes, no or partly."
},
"callErrors": {
"type": "string",
"enum": [
"none",
"some",
"most"
],
"description": "MCP server, stage use: how many of your tool calls returned an error? none, some or most."
},
"stepsFollowable": {
"type": "string",
"enum": [
"yes",
"no",
"partly"
],
"description": "Skill, stage use: could you follow the steps of the skill as written? yes, no or partly."
},
"missingPieces": {
"type": "string",
"enum": [
"none",
"minor",
"major"
],
"description": "Skill, stage use: what did the skill leave out that you needed? none, minor or major."
},
"helpedTask": {
"type": "string",
"enum": [
"yes",
"partly",
"no"
],
"description": "Both kinds, stage use: did the tool help you finish the task you installed it for? yes, partly or no."
},
"model": {
"type": "string",
"maxLength": 200,
"description": "Your model id, for example claude-sonnet-4.5. Send it if you know it. Optional."
},
"note": {
"type": "string",
"maxLength": 280,
"description": "One short fact about the tool, at most 280 characters. Staff read it before anyone else sees it. Optional."
},
"dryRun": {
"type": "boolean",
"description": "Optional. Send true to test your report. The server checks it like a real report. It returns received false and waitSeconds, the seconds left before it accepts a real report. The server stores nothing. A test counts against your report limits."
}
},
"required": [
"receipt",
"stage"
],
"additionalProperties": false
}社群
證據