integration · openai

Claude + OpenAI — run GPT tools natively.

Expose OpenAI's full suite of developer APIs—including chat completions, assistants, files, and vector stores—directly to your Claude Desktop, Claude Code, or LangChain agents via the open MCP standard.

Last updated 2026-05-28 · works with Claude, OpenAI, LangChain, and any MCP client

Positioning

vs. the alternatives

Capability	Raw OpenAI SDK	Composio OpenAI	wmcp.sh
Setup time	Install SDK + key configurations	Platform auth configurations	Sign in & provide key — done
Interoperability	Locked into OpenAI client	Custom wrapper	Standard MCP — works with Claude/Cursor
Spec updates	Wait for SDK release	Wait for platform update	Instant spec resolution at the edge
Latency	80-200ms SDK bootstrap	Platform router delays	Sub-50ms Edge proxy resolution
Tool Coverage	Manual method mapping	Curated subset	Every public endpoint out-of-the-box

What you get

Tools your agent calls

All MCP-shaped — consumed directly by Claude tool_use, OpenAI function-calling, or LangChain.

Tool	Type	Returns
`chat/completions`	Live action	GPT-4o chat completions JSON response
`embeddings/create`	Live action	Text embeddings vector coordinates
`assistants/create`	Live action	New Assistant instance details
`threads/create`	Live action	Assistant thread session ID
`files/create`	Live action	Uploaded file reference metadata

Integrate

Python — Claude calling OpenAI embeddings through wmcp.sh

from wmcp import WmcpClient
from wmcp.anthropic import to_anthropic_tools, execute_tool_use
from anthropic import Anthropic

client = WmcpClient(api_key="webmcp_live_…")
spec   = "https://raw.githubusercontent.com/openai/openai-openapi/master/openapi.yaml"
tools  = client.tools(spec)

# Filter to embeddings tools
embedding_tools = [t for t in tools if t.name.startswith("embeddings")]

anthropic = Anthropic()
msg = anthropic.messages.create(
    model="claude-3-5-sonnet-20241022",
    max_tokens=1024,
    tools=to_anthropic_tools(embedding_tools),
    messages=[{"role": "user",
               "content": "Vectorize the query: 'shopper side agent stack'."}],
)

for block in msg.content:
    if block.type == "tool_use":
        result = execute_tool_use(client, spec, block.model_dump())
        print(result)

FAQ

Frequently asked

Do I need a paid OpenAI key?

Yes. You provide your own OpenAI API key in your secure client header or vault. wmcp.sh does not provide API credits.

How does this differ from the official OpenAI SDK?

The SDK requires locking your codebase into OpenAI's client library. wmcp.sh maps the OpenAI OpenAPI spec to the universal MCP standard, letting Claude Desktop, Claude Code, and other editors run OpenAI actions natively.

Are my API keys secure?

Yes. wmcp.sh routes all requests through our secure PKCE credentials proxy. Your keys are encrypted out-of-band and never exposed in the LLM context.

API agent-readiness

OpenAI's chat completions is the multi-model orchestration target. If you're shipping your own model/API surface, see the API-side checklist.

→ /agent-ready/api · cornerstone · done-for-you ($499+)

Need this done for you?

Skip the wiring — we build, deploy, and monitor.

Custom adapter + hosted MCP at mcp.yourbrand.com + verified badge. From $499 one-time setup.

See /managed → Submit (free)