# Agno v3.1.9 adds an HTTP endpoint for running MCP tools

published: 2026-10-22T19:05:39.831Z
topic: frameworks
source: https://github.com/agno-agi/agno/releases/tag/v3.1.9
url: https://aiagentnews.io/stories/agno-v3-1-9-adds-an-http-endpoint-for-running-mcp-tools

Agno has released v3.1.9, which adds an HTTP route for running MCP tools listed on the AgentOS Server Card. The release also fixes bugs across the AgentOS MCP server, workflows, chunking, Postgres metrics, human-in-the-loop input and Mistral parameters.

The main addition is the MCP Tool Runner, which accepts POST /mcp/server/tools/{tool_name}/run. It executes any tool listed on the AgentOS Server Card over HTTP and returns the result, so a tool can be tested without an MCP client. The endpoint can be turned off with MCPConfig(tool_run_endpoint=False).

Several fixes affect the AgentOS MCP server. /mcp and /mcp/server-card now accept browser requests from origins that the AgentOS CORS policy allows, and /info reports the configured MCP path instead of always returning /mcp. Postgres metrics refresh on a migrated database no longer loads the legacy sessions.runs column into memory.

Other fixes cover function workflows, AgenticChunking, BaiduSearchTools and Mistral. A requires_user_input tool with a parameter typed int | None now pauses cleanly, and the paused run resumes with continue_run(run_id=...). The timeout, endpoint and safe_mode parameters on MistralChat now work, and safe_mode is deprecated in favor of safe_prompt.

## Why it matters if you build agents

Developers can now test an MCP tool with a plain HTTP call instead of setting up an MCP client first. Code that passes safe_mode to Mistral classes should move to safe_prompt, since the old name is deprecated. Teams exposing AgentOS to browsers should review the updated CORS behavior and the MCPConfig switch that disables the new endpoint.