Import jpcite into OpenAI Codex CLI / Agents SDK in one line

Five-minute setup

1. Install openai-agents: pip install openai-agents (Python 3.10+).
2. Build jpcite with HostedMCPTool: write mcp_tool = HostedMCPTool(tool_config={'type': 'mcp', 'server_label': 'jpcite', 'server_url': 'https://api.jpcite.com/mcp', 'require_approval': 'never'}) in a single line.
3. Bind it to an Agent: agent = Agent(name='jp_subsidy', tools=[mcp_tool]) binds all 261 MCP tools to the Agent.
4. (Optional) Pass an API key: add 'headers': {'X-API-Key': os.environ.get('JPCITE_API_KEY', '')} to tool_config to upgrade to metered mode. Omit it to stay on the anonymous free tier.
5. Verify: in the next section, confirm that Runner.run_sync(agent, input='...') answers through the 261 MCP tools.

Copy-paste snippet

A 30-line sample that defines the Agent and runs through a search_programs call.

import os
from agents import Agent, HostedMCPTool, Runner

# 1) Build the jpcite hosted MCP endpoint as a HostedMCPTool
mcp_tool = HostedMCPTool(
    tool_config={
        "type": "mcp",
        "server_label": "jpcite",
        "server_url": "https://api.jpcite.com/mcp",
        "require_approval": "never",
        "headers": {
            # On the anonymous free tier (3 req/IP/day) you can drop "headers" entirely
            "X-API-Key": os.environ.get("JPCITE_API_KEY", ""),
        },
    },
)

# 2) Bind the jpcite tools to an Agent
agent = Agent(
    name="jp_subsidy",
    instructions=(
        "You are an assistant for Japanese subsidies, regulations, and tax. "
        "Always call the jpcite tools and cite source_url and fetched_at in your answer. "
        "Do not provide individualized tax or legal advice (e.g., Certified Tax Accountant Act §52, "
        "Attorney Act §72); only cite general information and recommend consulting a professional."
    ),
    tools=[mcp_tool],
)

# 3) Run it
result = Runner.run_sync(
    agent,
    input="List 3 of the latest manufacturing subsidies in Tokyo (ものづくり補助金), with source_url",
)
print(result.final_output)

Verification

Send ものづくり補助金 東京都 through the Agent and confirm that the answer comes back via the jpcite tools with a source_url. There are 261 public tools; generation of 36-Agreement labor documents is not offered publicly.

python -c "import os; from agents import Agent, HostedMCPTool, Runner; \
t = HostedMCPTool(tool_config={'type':'mcp','server_label':'jpcite','server_url':'https://api.jpcite.com/mcp','require_approval':'never'}); \
a = Agent(name='probe', tools=[t]); \
print(Runner.run_sync(a, input='ものづくり補助金 東京都').final_output)"

Expected: a response that includes a subsidy name plus a source_url. Confirm the tools list (261) separately on the server side with curl -s https://api.jpcite.com/mcp/tools | jq 'length'.

When smoke-testing the REST endpoint directly, always use -G --data-urlencode. Embedding a Japanese query straight into the URL breaks the HTTP request line, and the API returns {"error":{"code":"bad_request",...,"hint":"Use --data-urlencode for non-ASCII query params..."}}.

curl -G "https://api.jpcite.com/v1/programs/search" \
  --data-urlencode "q=ものづくり補助金" \
  --data-urlencode "prefecture=東京都" \
  -H "X-API-Key: ${JPCITE_API_KEY:-}"

FAQ