langchain-openapi-tools 2.0.0

Convert OpenAPI specifications into native, production-grade LangChain tools.

langchain-openapi

Convert OpenAPI v3.0 & v3.1 and Swagger 2.0 specifications into native, production-grade LangChain tools.

---

langchain_openapi is a modern Python library designed to seamlessly convert OpenAPI v3.0, v3.1, and Swagger 2.0 specifications into native, type-safe LangChain tools for AI agents and LLM applications.

No python code generation is required—tools are generated dynamically at runtime with strict Pydantic input schemas, prompt optimization options, and production-ready middleware.

---

Installation & Importing

Install the PyPI package:

pip install langchain-openapi-tools
# or using uv
uv add langchain-openapi-tools

Import in Python:

from langchain_openapi_tools import OpenAPIToolkit, OpenAPIToolkitConfig

Migration Note

# Old (deprecated, but still works for backward compatibility):
from langchain_openapi import OpenAPIToolkit

# New (recommended):
from langchain_openapi_tools import OpenAPIToolkit

---

Features

• ⚡ Zero-Code Tool Generation: Runtime conversion of OpenAPI specs (YAML/JSON) into LangChain StructuredTools.
• 🚀 Swagger 2.0 & OpenAPI 3.x: Native normalization of legacy Swagger 2.0 and modern OpenAPI 3.0/3.1 specs.
• 🗜️ Prompt Optimization & Filtering: Description modes (full, compact, minimal), description compression, overrides, callbacks, and operation/tag filtering to drastically reduce context window usage.
• 🔒 Pluggable Authentication: Built-in support for Bearer Tokens, API Key Headers, Query Parameters, Basic Auth, and custom request providers.
• 🛡️ Production Middleware: Composable middleware architecture for Retries (exponential backoff), Rate Limiting (token-bucket), Caching (TTL), Pagination aggregation, and Sanitized Logging.
• 🎯 Type-Safe Validation: Dynamically generated Pydantic input schemas ensure LLM arguments adhere strictly to specification types before network transport.
• 🌐 Async Engine: Non-blocking asynchronous network transport powered by httpx.AsyncClient.

---

Supported Specifications

• ✅ Swagger 2.0 (Automatically normalized to OpenAPI 3.0 via the built-in adapter layer)
• ✅ OpenAPI 3.0.x (JSON and YAML)
• ✅ OpenAPI 3.1.x (JSON and YAML, including union type arrays, oneOf / anyOf / allOf, and const)

Note: Swagger 2.0 is commonly referred to as OpenAPI 2.0.

Compatibility Matrix

Feature	Swagger 2.0	OpenAPI 3.0	OpenAPI 3.1
Path / query / header / cookie params	✅	✅	✅
$ref resolution (internal)	✅	✅	✅
Base URL from host + basePath	✅	—	—
Base URL from servers[]	—	✅	✅
Fallback base URL from spec source URL	✅	✅	✅
application/json request bodies	✅	✅	✅
application/x-www-form-urlencoded	✅	✅	✅
multipart/form-data (incl. files)	✅	✅	✅
text/* and application/xml	✅	✅	✅
Vendor +json media types	✅	✅	✅
oneOf / anyOf (as Python Union)	—	✅	✅
allOf merge into single model	—	✅	✅
nullable: true (3.0)	—	✅	—
type: ["string", "null"] (3.1)	—	—	✅
const	—	—	✅
readOnly / writeOnly / deprecated	✅	✅	✅
Security: Bearer / API key / Basic	✅	✅	✅
Async execution (httpx.AsyncClient)	✅	✅	✅

Not yet supported: discriminator-driven union routing, patternProperties, $dynamicRef / $dynamicAnchor, callbacks, links, webhooks, and XML request-body serialization from Python dicts.

---

Quick Start

from langchain_openapi_tools import OpenAPIToolkit

# Load spec from remote URL or local file
toolkit = OpenAPIToolkit.from_url("https://api.crossref.org/swagger-docs")

# Extract generated tools
tools = toolkit.get_tools()

print(f"Generated {len(tools)} tools:")
for tool in tools[:3]:
    print(f"- {tool.name}: {tool.description}")

Agent Integration

import asyncio
from langchain_openapi_tools import OpenAPIToolkit


async def main():
    toolkit = OpenAPIToolkit.from_url("https://api.crossref.org/swagger-docs")
    search_tool = toolkit.get_tool("get_works")

    if search_tool:
        result = await search_tool.ainvoke({"query": "LangGraph"})
        print("Search Result:", result)


if __name__ == "__main__":
    asyncio.run(main())

---

Prompt Optimization & Tool Customization

Large OpenAPI specifications can generate extensive tool descriptions that exceed model context windows. langchain_openapi_tools provides full control over description generation and context footprint.

Configuration Object (OpenAPIToolkitConfig)

from langchain_openapi_tools import OpenAPIToolkit, OpenAPIToolkitConfig

config = OpenAPIToolkitConfig(
    description_mode="compact",
    compress_descriptions=True,
    include_tags=["Works"],
    tool_description_overrides={
        "get_works": "Search scholarly papers registered with Crossref."
    },
)

toolkit = OpenAPIToolkit.from_url(
    "https://api.crossref.org/swagger-docs", config=config
)

Description Modes (description_mode)

• full (default): Complete summary, description, HTTP method, path, and schema parameter details.
• compact: Summary and short parameter list without response schemas or redundant examples.
• minimal: A single-sentence summary ideal for large specifications with dozens of tools.

toolkit = OpenAPIToolkit.from_url(url, description_mode="minimal")

Description Compression (compress_descriptions=True)

Removes duplicate text, redundant whitespace, and empty sections without altering tool semantics:

toolkit = OpenAPIToolkit.from_url(url, compress_descriptions=True)

Custom Overrides & Callbacks

Override descriptions for specific tools:

toolkit = OpenAPIToolkit.from_url(
    url, tool_description_overrides={"get_works": "Search scholarly papers."}
)

Or pass a custom builder callback:

def my_builder(operation):
    return f"Execute {operation.name} on path {operation.path}."


toolkit = OpenAPIToolkit.from_url(url, description_builder=my_builder)

Operation Filtering

Filter tools before they are created to reduce context window overhead:

# Filter by Tags
toolkit = OpenAPIToolkit.from_url(url, include_tags=["Works"], exclude_tags=["Admin"])

# Filter by Operations
toolkit = OpenAPIToolkit.from_url(
    url, include_operations=["get_works"], exclude_operations=["delete_work"]
)

---

Architecture Pipeline

          Swagger 2.0 / OpenAPI 3.x
                     │
                     ▼
            Swagger Normalizer
                     │
                     ▼
          Normalized OpenAPI Model
                     │
                     ▼
              Existing Parser
                     │
                     ▼
           Internal Operation Models
                     │
                     ▼
            Schema Converter
                     │
                     ▼
              HTTP Executor
                     │
                     ▼
          LangChain StructuredTools

Example:

from langchain_openapi_tools import OpenAPIToolkit
from langchain.agents import create_agent
from langchain.chat_models import init_chat_model

toolkit = OpenAPIToolkit.from_url(
    "https://api.crossref.org/swagger-docs"
)

model = init_chat_model("openrouter:qwen/qwen3-32b:free")

agent = create_agent(
    model=model,
    tools=toolkit.get_tools(tags=["Works"]),
)

response = agent.invoke({
    "messages": "Find papers about Kashmir"
})