# Agno v3.1.9

Category: agent-infrastructure
Published: 2026-10-22T16:50:03Z
Source: [Agno releases](https://github.com/agno-agi/agno/releases/tag/v3.1.9)
Agent usefulness: 62/100
Confidence: 0.96
Content mode: release-watch
Verified: 2026-10-23T18:18:05.589Z
Tags: agno, agent-runtime, python

## Human Summary
Agno v3.1.9 introduces an HTTP endpoint for executing MCP tools directly without an MCP client, improves AgentOS MCP server CORS support, and fixes various bugs across chunking, metrics, and integrations.

## Agent Summary
Agno v3.1.9 adds POST /mcp/server/tools/{tool name}/run for testing MCP tools via HTTP (configurable via MCPConfig(tool run endpoint=False) ). Fixes include handling PEP 604 union types in HITL tools, resolving zero/negative breakpoint hangs in AgenticChunking , supporting sync functions/generators in async workflows, addressing Mistral client parameter TypeErrors, and correcting CORS handling on AgentOS MCP…

## Body
Agno v3.1.9 brings several improvements and bug fixes across the framework: - MCP Tool Runner : Added POST /mcp/server/tools/{tool name}/run , allowing any tool listed on the AgentOS Server Card to be executed over HTTP without needing an MCP client. This behavior can be disabled via MCPConfig(tool run endpoint=False) . - AgentOS MCP Server : /mcp and /mcp/server-card endpoints now properly respect AgentOS CORS policies for browser origins (enabling os.agno.com access to local instances). /info now correctly reports the configured MCP path. - Human-in-the-Loop : Tools marked with requires user input with union parameter types (such as int | None or X | Y ) now pause cleanly and resume properly using continue run(run id=...) . - AgenticChunking : Resolved an issue where chunking hung on zero or negative model breakpoints, and enforced validation that max chunk size must exceed 0. - Function Workflows : Workflow step functions raising TypeError within their body now run once instead of looping incorrectly, and sync functions/generators now properly operate with arun() . - Mistral Integration : Fixed TypeError regressions when passing timeout , endpoint , and safe mode in MistralChat and endpoint in MistralEmbedder . safe mode is now deprecated in favor of safe prompt . - Postgres Metrics : Prevented memory spikes during metrics refreshes by no longer loading the legacy sessions.runs column into memory. - BaiduSearchTools : Fixed encoding issues causing Chinese titles and abstracts to display as \uXXXX escape sequences.

## Recommended actions
- Migrate safe mode usage in MistralChat to safe prompt .
- Disable the MCP HTTP tool runner using MCPConfig(tool run endpoint=False) if direct HTTP invocation of tools without an MCP client should not be exposed.

## Sponsors
No sponsor placement attached.

## Agent-readable Sponsor Surface
Sponsor inventory is available at /api/sponsors.json with useCases, pricing, API/docs URLs, targetAgents, constraints, CTA URL, commercial disclosure fields, sourceOfTruthUrl, constraintsLastVerifiedAt, constraintsRefreshCadence, driftHandlingPolicy, and constraintPolicy.