AGENT ORCHESTRATION & MCP FRAMEWORKS

pydantic-ai vs mcp-agent

Security and governance comparison between Pydantic-AI and LastMile AI mcp-agent across static execution sinks and deterministic lockfile hygiene.

lastmile-ai/mcp-agent leads this category by +10 points in the MSHB-2026 6-vector security benchmark.
Run Live Repo Audit →
pydantic/pydantic-ai
★ 21,014·SHA: 5616c219·MIT
MSHB SCORE86/100Grade A

Governance & Supply-Chain Hygiene

Missing `SECURITY.md` vulnerability disclosure policy

Static Execution Sinks (2)

[MEDIUM] Sensitive Credential / Full-Env Access.github/set_docs_main_preview_url.py:L9
GITHUB_TOKEN = os.environ['GITHUB_TOKEN']
[LOW] External Network Egress Call.github/set_docs_main_preview_url.py:L44
r = httpx.post(deployment_url, headers=gh_headers, json=deployment_data)
lastmile-ai/mcp-agent
★ 8,575·SHA: f62d8493·Apache-2.0
MSHB SCORE96/100Grade A+

Governance & Supply-Chain Hygiene

Full governance hygiene verified (SECURITY.md, lockfile, license).

Static Execution Sinks (1)

[MEDIUM] Sensitive Credential / Full-Env Accesssrc/mcp_agent/app.py:L267
os.environ[key] = str_value

More Head-to-Head MCP Security Comparisons