OpenAI's official agent framework, pointed at every social platform you run.

OpenAI shipped the Agents SDK in 2025 as the official Python framework for building agentic apps on OpenAI models. It supports MCP servers natively — no third-party adapter — through MCPServerStreamableHttp for HTTP transports like SociaHive's at https://sociahive.com/api/mcp. Once registered, every SociaHive tool flows in as an SDK Tool the agent can call, with the same tracing, hand-off, and guardrail primitives the SDK provides for other tool sources.

Why pair OpenAI Agents SDK with SociaHive

Register SociaHive with MCPServerStreamableHttp

The Agents SDK's MCPServerStreamableHttp class connects to any HTTP-transport MCP server. Pass SociaHive's URL plus a headers dict with X-API-Key. The Agent picks up SociaHive's tools at construction time.

# pip install openai-agents
from agents import Agent, Runner
from agents.mcp import MCPServerStreamableHttp
import os

sociahive = MCPServerStreamableHttp(
    name="sociahive",
    params={
        "url": "https://sociahive.com/api/mcp",
        "headers": {
            "X-API-Key": os.environ["SOCIAHIVE_API_KEY"],
        },
    },
)

await sociahive.connect()   # discover tools

agent = Agent(
    name="Social ops",
    instructions=(
        "You manage cross-platform social presence via SociaHive. "
        "Use schedule_post for content; generate_flow for DM automations. "
        "Surface pending_confirmation envelopes to the human; never auto-approve."
    ),
    mcp_servers=[sociahive],
)

result = await Runner.run(
    agent,
    input="Pull last 7d Instagram + LinkedIn analytics, draft a Threads recap, "
          "and schedule it for Sunday 6pm. Stop at confirmation.",
)
print(result.final_output)

A real workflow: scheduled weekly recap with hand-off to a reviewer agent

One agent pulls analytics and drafts the recap; a second agent (with a different system prompt and a publish-rights API key) reviews and clears pending_confirmation envelopes. The hand-off happens automatically via the SDK's hand-off primitive.

1. 1Drafter agent reads its instructions ('pull analytics, draft a recap, schedule but never publish'). It's wired with a read+write SociaHive key but its system prompt explicitly forbids resolving confirmations.
2. 2Drafter calls get_post_analytics for each connected platform via the SociaHive MCP server. The SDK's tracing surface logs every call.
3. 3Drafter calls create_post twice — Threads short, LinkedIn longform — both with scheduledAt = next Sunday 18:00. Both return pending_confirmation envelopes.
4. 4Drafter invokes hand_off to the Reviewer agent, passing the envelopes as structured input. The Reviewer agent has a different SociaHive API key (one with confirmation-resolution rights).
5. 5Reviewer reads the drafts, evaluates against brand-voice guardrails, and either resolves both confirmations through the SociaHive dashboard URL the envelopes carry, or returns the run to the Drafter with edit notes.

from agents import Agent, handoff, Runner
from agents.mcp import MCPServerStreamableHttp
import os

drafter_mcp = MCPServerStreamableHttp(
    name="sociahive_drafter",
    params={"url": "https://sociahive.com/api/mcp",
            "headers": {"X-API-Key": os.environ["SOCIAHIVE_DRAFTER_KEY"]}},
)
reviewer_mcp = MCPServerStreamableHttp(
    name="sociahive_reviewer",
    params={"url": "https://sociahive.com/api/mcp",
            "headers": {"X-API-Key": os.environ["SOCIAHIVE_REVIEWER_KEY"]}},
)

await drafter_mcp.connect()
await reviewer_mcp.connect()

reviewer = Agent(
    name="Reviewer",
    instructions=(
        "You receive pending_confirmation envelopes from the Drafter. "
        "Approve or request changes. Never bypass the confirmation step."
    ),
    mcp_servers=[reviewer_mcp],
)

drafter = Agent(
    name="Drafter",
    instructions=(
        "Pull analytics and draft scheduled posts. Hand off pending "
        "confirmations to the Reviewer — never resolve them yourself."
    ),
    mcp_servers=[drafter_mcp],
    handoffs=[handoff(reviewer)],
)

result = await Runner.run(drafter, input="weekly recap, sun 6pm")

What You Can Do with OpenAI Agents SDK + SociaHive

• ✓Native MCP support — no third-party adapter, just the openai-agents Python package
• ✓Connect once with MCPServerStreamableHttp; every SociaHive tool appears in the agent automatically
• ✓Built-in tracing in OpenAI's Traces dashboard alongside your model and tool calls
• ✓Hand-offs and guardrails work the same against SociaHive as against any other agent tool

How to Connect OpenAI Agents SDK

Use Cases with OpenAI Agents SDK

Other AI Agent Integrations

Frequently asked questions