Opengos

What opengos MCP server does

The opengos MCP server provides an agent-facing workflow for grants and public-goods funding. Its stated responsibilities are discovering opportunities, ranking potential matches, and producing grounded draft materials. The project presents these stages as a discoverer, ranker, and drafter sequence rather than as an autonomous approval process.

OpenGOS follows a proposal-and-review model. Agents can generate candidate matches and drafts, but a human or kernel reviews a proposal before it is sent. Ranked grant matches and draft artifacts are treated as established outputs; an agent's own claim of success is not treated as verification.

How it works

The opengos MCP server exposes the grants workflow through MCP. The multi-agent implementation separates discovery, ranking, and drafting into successive stages, with the workflow materials located under skills/multi-agent-workflow/ in the repository.

The surrounding FOUNDRY design separates untrusted proposal generation from host-owned verification. In that model, the host owns the evaluator, while agents propose. The README also identifies YodMCP as the durable-memory location for established facts and describes promotion through LRSI, Server-OS, and RUI as fail-closed. These related components are context for the project architecture, not requirements stated for the basic OpenGOS quick start.

Setup and configuration

The repository identifies Python 3.11 or newer and provides a Python SDK. The documented development installation is:

pip install -e ".[dev,api]"

After installation, the main entry points are:

opengos
opengos-cli status
opengos-api

opengos starts the MCP surface. opengos-cli status checks the CLI surface, while opengos-api starts the REST service on port 8080 according to the README. The API documentation is available at http://localhost:8080/docs when that service is running. No required environment variables or external credentials are documented in the provided material.

Tools and capabilities

The documented capabilities include:

• Grant and public-goods opportunity discovery
• Ranking grant matches
• Grounded drafting of proposal materials
• A command-line interface with status and search commands
• A Python client available as OpenGOSClient
• A FastAPI-based REST API exposed through opengos-api

The README also includes an interactive HTML demo and identifies continuous integration configuration, but it does not provide a complete list of individual MCP tool names or request schemas.

Limitations and notes

The opengos MCP server should be treated as a proposal-generation and drafting aid, not as an approval or submission authority. Human or kernel review is expected before a proposal is sent. The supplied material does not document grant-source connectors, authentication settings, environment variables, individual MCP tool schemas, or specific grant databases, so those details should be confirmed in the repository before integration.

The project is licensed under Apache-2.0. The quick start uses an editable installation with development and API extras, which may be more than is needed for an MCP-only deployment; the excerpt does not specify a smaller installation command.