Cross-Platform MCP Server for Gopher and Gemini

Overview

This project provides a cross-platform MCP (Model Context Protocol) server that lets AI assistants safely browse, fetch, and interact with resources on the Gopher and Gemini networks. It exposes protocol-specific capabilities as MCP-compatible tools so language models can request content, follow links, and obtain rendered previews in a controlled, auditable fashion.

Running an MCP server for Gopher and Gemini centralizes policy, sanitization, and resource limits for browsing operations. Instead of letting a model open arbitrary network connections, you point it at this server and grant access only to well-defined tool actions (fetch, list, search, render). That improves safety and reproducibility while enabling assistants to use niche internet protocols that most modern clients no longer support.

Features

• Protocol support for Gopher and Gemini (fetching, listing, and rendering)
• MCP-compatible tool endpoints so models can call browsing actions
• Content sanitization and size limits to prevent abusive responses
• Configurable TLS support and host allowlists
• Docker and source build options for cross-platform deployment
• Lightweight, low-dependency runtime suitable for edge or cloud instances

Installation / Configuration

Clone the repository and run from source or use Docker. The repository is available at: https://github.com/cameronrye/gopher-mcp

Build from source (generic steps):

git clone https://github.com/cameronrye/gopher-mcp.git
cd gopher-mcp
# If the project is implemented in Go:
go build -o gopher-mcp ./cmd/server
# or run directly if using a language/runtime
./gopher-mcp --config ./config.yml

Run with Docker:

docker pull ghcr.io/cameronrye/gopher-mcp:latest
docker run -p 8080:8080 \
  -v $(pwd)/config.yml:/app/config.yml:ro \
  ghcr.io/cameronrye/gopher-mcp:latest \
  --config /app/config.yml

Example minimal YAML configuration:

server:
  listen: "0.0.0.0:8080"
  tls:
    cert_file: ""
    key_file: ""
mcp:
  service_name: "gopher-mcp"
  max_response_bytes: 262144
  request_timeout_seconds: 15
network:
  user_agent: "gopher-mcp/1.0"
  allowed_hosts: ["gopher.floodgap.com", "*.gemini.circumlunar.space"]
logging:
  level: "info"

Common configuration options

Adjust these values to match your deployment policies (for example, stricter timeouts and smaller response limits for publicly exposed instances).

Available Tools / Resources

The server exposes a set of MCP tools representing common browsing actions. Typical tool names and behaviors include:

• gopher.fetch — fetch a Gopher selector and return raw or parsed content
• gopher.list — list a Gopher directory (menu) entries
• gemini.fetch — fetch a Gemini URL and return raw bytes and metadata
• gemini.render — produce a sanitized, text/HTML preview suitable for model consumption
• search.index — optional indexing tool to provide keyword searches over fetched content
• link.preview — follow a link with safelisting and produce a short summary and content snippet

Each tool returns structured JSON-like results with status codes, MIME/type metadata, and truncated content to keep model context bounded. Refer to the MCP tool schema (embedded in the repo) for exact field names and error formats.

Use Cases

• Assistants that need to answer questions about legacy or niche content: an LLM can call gopher.list to enumerate a Gopher site and gopher.fetch to retrieve a specific document, then synthesize an answer without direct outbound network access.
• Content discovery and summarization: use gemini.fetch + gemini.render to retrieve Gemini capsules, convert them into sanitized text, and feed summaries back into a model for downstream analysis or indexing.
• Controlled browsing for multi-step tasks: chain MCP tool calls (search.index -> gopher.fetch -> link.preview) to let a model research and cite sources while enforcing size and host restrictions.
• Integration into pipelines: run the server behind an internal firewall and expose only the MCP surface to downstream agents or services, centralizing policy for TLS, timeouts, and logging.

Getting Started Tips

• Start with conservative limits (small max_response_bytes, short timeouts) and expand only after monitoring usage.
• Maintain an allowlist of trusted hosts rather than a broad allow-any policy for public instances.
• Enable TLS when exposing the server beyond a trusted network.
• Capture tool call logs for auditing model behavior and debugging agent interactions.

For full implementation details, usage examples, and schema references, consult the repository at https://github.com/cameronrye/gopher-mcp.