A second set of eyes inside your agent

Software Sausage MCP + Overwatch

Find bounded agent workflows and commercial or open-source software for free. Members can send an agent's proposed answer, plan, or artifact to a different model for a critique before acting on it.

PUBLIC DEMO

Connect first, pay only if useful

The demo includes read-only search_software and find_agent_recipes tools plus complete recipe resources, limited to five tool calls per minute per network address. No account or card.

https://softwaresausage.com/api/mcp

OVERWATCH · $10/MONTH

Challenge agent output before you trust it

Includes the same recipe and catalog tools plus up to 100 review_agent_output critiques each month. Qwen 3.8 27B checks unsupported claims, ambiguity, omissions, and what to verify next. Thirty total tool calls per minute, one rotatable bearer token, and Stripe-hosted cancellation.

Overwatch is a critique, not ground truth. It does not watch sessions continuously, browse sources, execute code, or change your agents.

Connect the free demo in one command

Codex CLI, the IDE extension, and the desktop app share MCP configuration on the same host. Add the public Streamable HTTP endpoint, then restart your client.

codex mcp add software_sausage --url https://softwaresausage.com/api/mcp

Try: Use Software Sausage to find an agent workflow for reviewing an AI-authored code change.

Member setup after subscribing

Create the token above, export it in the shell that starts Codex, and keep it out of repositories:

export SOFTWARE_SAUSAGE_MCP_TOKEN='ss_mcp_…'

Add this server to ~/.codex/config.toml:

[mcp_servers.software_sausage]
url = "https://softwaresausage.com/api/mcp/member"
bearer_token_env_var = "SOFTWARE_SAUSAGE_MCP_TOKEN"

Ask your agent: Use review_agent_output to challenge this plan before I run it.

Read the official Codex MCP configuration guide.

What the tools return

find_agent_recipes: up to five job matches with roles, expected artifact, safety boundary, canonical link, and a read-only sausage://recipes/{slug} resource.
search_software: up to five commercial entries and five open-source alternatives with operating tradeoffs and source links.
review_agent_output: a verdict, concise summary, up to five concerns, and up to five checks to run next.
Model and run-time disclosure on every Overwatch critique. Submitted text is not stored in D1.

This is decision support, not procurement approval. Pricing, licensing, security, and feature claims change; verify them with the vendor or project before purchase or deployment.

Read the editorial methodology and privacy policy.