Add example: external HTTP MCP server (TensorFeed.ai)

examples/external_mcp_tensorfeed.py

@@ -0,0 +1,82 @@
+#!/usr/bin/env python3
+"""Example: External hosted MCP server (TensorFeed.ai cross-database CVE verification).
+
+This example demonstrates how to wire an external HTTP MCP server into
+the Claude Agent SDK. The TensorFeed MCP server at https://tensorfeed.ai/api/mcp
+exposes 17 free tools across AI news, model pricing, AI service status,
+security advisories (MITRE CVE / CISA KEV / FIRST.org EPSS / OSV), SEC EDGAR
+filings, FDA regulatory data, and US energy indicators. No auth required.
+
+The demo asks Claude to verify CVE-2024-3094 (the XZ backdoor) across three
+independent vulnerability databases. The model autonomously sequences the
+calls (MITRE for the canonical record, CISA KEV for active exploitation
+status, FIRST.org EPSS for exploitation likelihood) and surfaces a
+confirmed_by list so the user can audit which sources backed the answer.
+
+The premise: the actual production failure mode for security agents is not
+hallucination but acting on a single source. Cross-source corroboration is
+the fix. The TensorFeed MCP server makes that one call in the agent loop
+instead of N parallel API integrations.
+
+Other endpoints worth trying (free, no auth):
+  - "What's the latest AI news about Anthropic?" -> get_news_articles
+  - "Is Claude up right now?" -> get_status_summary
+  - "Compare Opus and GPT-5 pricing" -> get_models
+  - "Find recent FDA Class I drug recalls" -> query_fda_drug_recalls
+  - "Search SEC EDGAR for Anthropic filings" -> search_sec_edgar
+"""
+
+import anyio
+
+from claude_agent_sdk import (
+    AssistantMessage,
+    ClaudeAgentOptions,
+    ResultMessage,
+    TextBlock,
+    query,
+)
+
+
+async def main() -> None:
+    options = ClaudeAgentOptions(
+        mcp_servers={
+            "tensorfeed": {
+                "type": "http",
+                "url": "https://tensorfeed.ai/api/mcp",
+            }
+        },
+        # Restrict to the security tools for this demo. Drop allowed_tools
+        # to expose all 17 TensorFeed tools to the model.
+        allowed_tools=[
+            "mcp__tensorfeed__get_cve_record",
+            "mcp__tensorfeed__get_kev_catalog",
+            "mcp__tensorfeed__get_epss_score",
+        ],
+        max_turns=8,
+    )
+
+    prompt = (
+        "Verify CVE-2024-3094 across multiple databases via the tensorfeed MCP server. "
+        "Call get_cve_record for the MITRE record, get_kev_catalog to check for active "
+        "exploitation, and get_epss_score for the FIRST.org probability. Then summarize: "
+        "severity_band, exploited_in_wild boolean (true if KEV has the CVE), "
+        "epss_probability, a confirmed_by list of databases that returned data, and a "
+        "one-sentence triage recommendation."
+    )
+
+    print(f"Asking Claude: {prompt}\n")
+    print("=" * 70)
+
+    async for message in query(prompt=prompt, options=options):
+        if isinstance(message, AssistantMessage):
+            for block in message.content:
+                if isinstance(block, TextBlock):
+                    print(block.text)
+        elif isinstance(message, ResultMessage):
+            if message.total_cost_usd:
+                print("=" * 70)
+                print(f"\nCost: ${message.total_cost_usd:.4f}")
+
+
+if __name__ == "__main__":
+    anyio.run(main)