RealWorldAPIs

com.realworldapis由 realworldapis.com 发布官方发布者v3.0.1更新于 Oct 25, 2026

Evidence-first registry of real-world APIs for AI agents, with verified metadata and comparison.

已验证Streamable HTTP可网页运行网页搜索与抓取开发者工具

概览

AI 生成的概览

一个远程只读的真实世界 API 注册表,让助手为物理世界任务查找、筛选、验证并比较 API。

可以做什么
针对开放式自然语言任务或目标推荐 API,候选清单来自托管推荐服务,比较仅限这些候选。按类别、能力、协议状态或就绪度搜索和筛选 API 目录。直接搜索已验证的能力图谱,查看任务级能力及其执行条件。验证某个已知 API 是否满足某一具体需求,返回 YES、NO、PARTIAL 或 UNKNOWN 及证据。按名称或 slug 获取单个 API 的完整决策档案,并对 2 至 8 个已知 API 逐字段比较。
适用场景
适合需要在真实世界 API(天气、地图、出行、物流、物联网、能源、机器人)之间做选择,或在集成前确认某个 API 是否满足特定需求的场景。也适合要求证据支撑、未知项不得当作已确认支持的选型工作。
运行要求
远程 Streamable HTTP 端点;目前无需 API 密钥,也未声明环境变量或请求头。客户端需支持远程 Streamable HTTP 连接,并需能访问该托管服务的网络。
安装前请注意
六个工具均为只读,仅返回目录与证据数据,不写入、不付款。结果来自托管的自有服务,可能独立于公开文档而变化,README 建议在生产集成前核对提供商文档。Unknown 值不代表已确认支持。

安装

在 SourceWeft 中

  1. 打开 控制台中的 RealWorldAPIs,将其添加到工作区。
  2. 为需要使用其工具的对话启用该服务。

Web executable,通过 Streamable HTTP。 远程服务在工作区中配置后即可从网页运行时运行。

交给智能体安装

把这段提示词粘贴到 Claude Code、Codex、Cursor 等智能体,或 SourceWeft 对话中。智能体会读取该服务器的安装指南,先向你展示发布者、端点和登录方式,经你同意后再把它添加到你的客户端。密钥不会写进任何命令。

阅读 https://sourceweft.com/mcp/com.realworldapis%2Frealworldapis/install.md,并按其中说明把这个 MCP 服务器接入我的客户端。修改任何配置之前,先告诉我它的发布者、端点和登录方式,等我确认。不要把 API 密钥写进命令或我会提交的文件;问我密钥应从哪里获取。

查看智能体所依据的安装指南

其他 MCP 客户端

把它添加到你客户端的 mcpServers 配置中。

{
  "mcpServers": {
    "realworldapis": {
      "type": "http",
      "url": "https://mcp.realworldapis.com/mcp"
    }
  }
}

README

RealWorldAPIs MCP

RealWorldAPIs is an evidence-first API registry for AI agents connecting to the physical world.

A real-world API connects software or AI agents to data, devices, infrastructure or services representing the physical world — including weather, maps, mobility, logistics, IoT, energy and robotics.

This repository documents and provides a publicly inspectable MCP interface implementation for RealWorldAPIs.

The live hosted RealWorldAPIs service is the source of truth for current tools, schemas, recommendation behavior and evidence semantics. The hosted service may evolve independently from the code visible in this repository.

One core evidence rule applies throughout the service:

Unknown remains Unknown and is never treated as confirmed support.

Public interface, proprietary intelligence

RealWorldAPIs intentionally separates its public interoperability surface from its proprietary decision systems.

Public / interoperability surface

The following are intended to be publicly discoverable and usable for interoperability:

  • the remote MCP endpoint and tool schemas
  • public HTTP endpoints and OpenAPI descriptions
  • public API catalog and capability/evidence outputs
  • public documentation and usage examples
  • factual provider metadata and evidence links
  • structured recommendation and verification results returned by the hosted service

Proprietary hosted systems

The following are not published or licensed through this repository:

  • the canonical RealWorldAPIs recommendation and ranking engine
  • natural-language task parsing and task-frame logic
  • capability retrieval, fusion and sibling-capability reranking logic
  • internal scoring rules, weights and optimization logic
  • verification and evidence-processing implementation
  • assertion-generation and research pipelines
  • benchmark tooling, active private evaluation sets and holdout sets
  • internal administration, proposal and approval systems

The MCP layer calls hosted canonical RealWorldAPIs services for recommendation and evidence-backed decision support. Publishing an interface, schema or result does not publish the underlying hosted implementation.

See PROPRIETARY_NOTICE.md for the repository rights notice.

Live service

No RealWorldAPIs API key is currently required to discover or call the public MCP tools.

MCP tools

The live RealWorldAPIs MCP service currently exposes six read-only tools:

ToolPurpose
find_api_for_taskNatural-language API selection using the canonical hosted recommender, followed by constrained comparison over returned candidates.
search_apisDeterministic search and filtering over canonical API fields.
search_capabilitiesSearch the verified capability graph directly.
verify_requirementEvidence-backed YES / NO / PARTIAL / UNKNOWN verification of one API against a specific requirement.
get_apiRetrieve one canonical API decision profile by slug or name.
compare_apisDeterministic side-by-side comparison of canonical fields.

The live tool set and schemas are authoritative. The source-visible implementation in this repository may lag the hosted service.

Which tool should I use?

If you need...Use
Recommendations for an open-ended natural-language task or desired outcomefind_api_for_task
Exact filtering by category, capability, protocol status or readinesssearch_apis
Direct discovery over verified task-level capabilitiessearch_capabilities
Evidence-backed adjudication of an exact API requirementverify_requirement
The canonical record for one already-known APIget_api
A field-by-field comparison of already-known candidatescompare_apis

find_api_for_task

Use when the user describes a goal rather than a specific API.

Example task:

json
{  "task": "I need my agent to connect to a car and predict weather on the route"}

The structured shortlist returned by the canonical hosted recommender is the source of truth. Any AI comparison is constrained to that shortlist and is not permitted to introduce outside APIs.

search_apis

Use when the task is explicit enough to filter the registry directly.

Example:

json
{  "categories": ["Geospatial"],  "agent_capabilities": ["locate"],  "mcp": "Yes",  "min_readiness": 60,  "limit": 10}

Only explicit stored values are used. For example, mcp: "Yes" must not be satisfied by an API whose MCP status is Unknown.

search_capabilities

Use when the task is better expressed as an exact capability or execution requirement than as a broad API category.

Capability results can include execution, access, approval, hardware, sandbox, physical-consequence and reversibility fields when available from the canonical evidence graph.

verify_requirement

Use when you need to adjudicate one API against one exact requirement.

Example:

json
{  "api": "here-location-services",  "requirement": "truck routing in Sweden"}

The verifier returns an evidence-backed YES, NO, PARTIAL or UNKNOWN result. Unknown is never silently promoted into support.

get_api

Example:

json
{  "identifier": "geoapify"}

The response is the canonical RealWorldAPIs decision profile, including provider metadata, capabilities, authentication, protocol status and evidence links.

compare_apis

Example:

json
{  "apis": ["Geoapify", "Google Maps Platform", "Mapbox"],  "fields": [    "category",    "capabilities",    "agent_capabilities",    "auth_methods",    "openapi",    "mcp",    "agent_readiness_score"  ]}

The comparison is deterministic. RealWorldAPIs does not infer missing support or silently replace Unknown with a guess.

Connect

Any MCP client that supports remote Streamable HTTP can connect to:

text
https://mcp.realworldapis.com/mcp

Example MCP configuration:

json
{  "mcpServers": {    "realworldapis": {      "url": "https://mcp.realworldapis.com/mcp"    }  }}

Codex-style TOML

toml
[mcp_servers.realworldapis]url = "https://mcp.realworldapis.com/mcp"

Evidence model

RealWorldAPIs is designed around a normalized evidence layer rather than free-form API recommendations.

The canonical dataset tracks fields such as:

  • provider and category
  • verified task-level capabilities
  • agent action primitives
  • authentication methods
  • OpenAPI
  • MCP
  • A2A
  • x402
  • llms.txt
  • public documentation status
  • authentication verification
  • access and setup requirements
  • approval requirements
  • hardware dependencies
  • sandbox availability
  • physical consequence
  • irreversibility
  • direct evidence links

Grounding rules

  1. The canonical structured dataset is the source of truth.
  2. Unknown is distinct from No.
  3. AI comparison may not introduce APIs outside the structured shortlist.
  4. Search and comparison tools use canonical stored evidence.
  5. Hard provider facts require the appropriate evidence quality under the hosted RealWorldAPIs evidence policy.
  6. Provider documentation should be checked before production integration.

Why RealWorldAPIs exists

General API directories are useful for broad discovery, but AI agents increasingly need a normalized way to answer questions such as:

  • Which exact API capability can perform this task?
  • Which APIs can locate a vehicle and route a technician?
  • Which weather APIs are both agent-actionable and machine-readable?
  • Which robotics APIs expose machine-usable interfaces?
  • Which APIs can be combined for a physical-world workflow?
  • What access, approval, hardware or safety constraints apply?
  • What evidence gaps remain before an agent should rely on a provider?

RealWorldAPIs focuses specifically on APIs that connect software and agents to the physical world.

Research and dataset

The public research page, State of Agent-Ready Real-World APIs 2026, summarizes the catalog snapshot across agent-readiness, protocol support and category-level evidence:

https://realworldapis.com/research/state-of-agent-ready-apis-2026

Machine-readable research data:

https://realworldapis.com/research/state-of-agent-ready-apis-2026.json

The public catalog is available at:

https://realworldapis.com/api/catalog.json

The verified capability graph is available at:

https://realworldapis.com/api/capabilities.json

Architecture

The MCP server is intentionally an interface layer over canonical hosted RealWorldAPIs services.

text
MCP client   |   +-- find_api_for_task --------> hosted canonical recommender   +-- verify_requirement --------> hosted evidence/verifier layer   +-- search_capabilities -------> canonical capability graph   +-- search_apis ---------------> canonical API catalog   +-- get_api -------------------> canonical API decision profile   +-- compare_apis --------------> canonical evidence fields

This avoids maintaining a second independent recommendation, ranking or evidence system inside the public MCP interface.

The implementation of the canonical hosted recommendation, ranking, verification and evidence-processing systems is proprietary and is not contained in this repository.

See docs/ARCHITECTURE.md for additional public interface details.

Repository layout

text
src/  worker.js               Publicly inspectable MCP interface implementation
examples/  requests.md             Example tool calls and expected semantics
docs/  ARCHITECTURE.md         Public data flow, grounding and failure behavior  DATA_MODEL.md           Important catalog fields and evidence semantics
CHANGELOG.md               Release historyCONTRIBUTING.md            Contribution and correction workflowPROPRIETARY_NOTICE.md      Repository rights and proprietary-service boundary

Status

The hosted RealWorldAPIs service evolves independently from this public repository.

For current production behavior, use the live MCP endpoint and canonical website endpoints. Third-party directories and this repository may temporarily show an older tool count, description, implementation detail or schema after a production deployment.

Contributing

Corrections to public API metadata are welcome when supported by current provider documentation. Please include the evidence URL and identify the exact field that should change.

Public contributions do not grant access to, or imply publication of, proprietary RealWorldAPIs ranking, verification, optimization, benchmark or research systems.

See CONTRIBUTING.md.

Rights and licensing

This repository is publicly source-visible but is not released under an open-source license.

No MIT, Apache, BSD, GPL or other open-source license is granted by this repository.

Use of the hosted RealWorldAPIs service is separate from rights in repository source code. Public protocol descriptions, schemas and factual outputs may be used for interoperability as permitted by applicable terms and law, but the proprietary hosted RealWorldAPIs systems are not licensed through this repository.

See PROPRIETARY_NOTICE.md.

Copyright © 2026 RealWorldAPIs. All rights reserved.

来源:README.md,提交 39545da

工具

6

find_api_for_task

Find API for Task

Find the best APIs for a natural-language physical-world task. Use this for open-ended task or outcome-based recommendations; use search_apis when you already know exact filters, get_api for one known API, or compare_apis for known candidates. Returns an authoritative canonical structured_results shortlist plus a best-effort AI comparison restricted to that shortlist. The canonical engine now applies Verified/High capability evidence before ranking: explicit hard requirements expressed with terms such as must, required, no or without can exclude a candidate only when verified evidence contradicts them. Unknown evidence remains eligible and is surfaced as uncertainty. requirements.mcp, openapi and auth are translated into explicit requirements; categories and capabilities remain task-shaping hints. free_tier and min_readiness are accepted for compatibility but are not hard filters.

输入参数task必填limitrequirements

search_apis

Search APIs

Search the canonical registry when you already know keywords or exact requirements. All supplied filter groups combine with AND logic; array capability filters require every supplied value, and protocol fields match the exact Yes/No/Unknown status, so Unknown never satisfies Yes. Results include each API's capability_summary and direct decision-profile URL. Results are sorted by agent_readiness_score descending and then API name, with 1–20 results returned (default 10). Use search_capabilities when you need capability-level filters such as approval, physical consequence, sandbox or access friction.

输入参数a2amcpx402limitqueryopenapillms_txtcategoriescapabilitiesmin_readinessagent_actionableagent_capabilities

search_capabilities

Search Verified Capabilities

Search the canonical 770-capability Verified/High decision graph directly. Use this when the question is about what an agent can actually do and under what execution conditions: task/capability text, API, category, action type, acted-on object, human approval, physical consequence, irreversibility, sandbox, hardware dependency, access friction or setup. All supplied filter groups combine with AND logic; multiple values inside one array filter are alternatives (OR). Unknown remains Unknown and may be filtered explicitly by passing the literal value Unknown.

输入参数limitqueryobjectsandboxapi_slugscategoriesprimitivesagent_setupaction_typesactionabilityhuman_approvalaccess_friction另有 3 个

verify_requirement

Verify API Requirement

Verify whether one specific API can satisfy one exact requirement using the canonical Verified/High capability graph plus linked assertion evidence. Returns YES, NO, PARTIAL or UNKNOWN with matched capability, access/setup prerequisites, approval, safety, reversibility, reliability, coverage and other requested dimensions; direct provider facts remain distinct from inferred policy assertions. Unknown remains Unknown.

输入参数api必填locationrequirement必填

get_api

Get API

Retrieve one canonical API decision profile when you already know its name or slug. The response includes the full Verified/High capability profile with execution, access, approval, reversibility, physical-consequence and evidence fields. Exact slug or exact API name is preferred; a partial identifier is accepted only when it uniquely resolves to one record.

输入参数identifier必填

compare_apis

Compare APIs

Compare 2–8 already-known APIs side by side using canonical catalog fields plus capability-graph summary fields such as verified capability count, physical consequences, approval relevance and irreversibility. Use get_api for the full decision profile of one API and search_capabilities for capability-level filtering. No missing value is inferred and Unknown remains Unknown.

输入参数apis必填fields

版本历史

2
  1. v3.0.1最新Sep 27, 2026
  2. v1.2.1Sep 16, 2026

评分

已满足适用检查项的 90%

该评分衡量此收录条目的完整程度,依据注册中心条目、端点与公开来源计算得出。它不是对服务器本身的评价,也不表示服务器安全或当前可用:可连通性是带日期的检查结果,评分也不包含安装量或用户评价。

  • 近 30 天内可连通必需25 分
  • 已声明认证方式必需10 分
  • 已列出工具必需15 分
  • 描述必需15 分
  • 图标10 分
  • 文档15 分
  • 许可证10 分

    请在仓库或注册中心条目中声明许可证。