Send Email

mcp.build

mcp.build is a community to build open source MCPs (Model Context Protocol servers) that adopted the Hal9 open source agent structure for its ease of use for deployment and interoperability.

Website: hal9ai.github.io/mcp.build (GitHub Pages from /docs)

This repo is structured so coding agents (Claude Code, Grok Build, Cursor, Codex, …) can contribute with minimal guesswork. Start at AGENTS.md and docs/llms.txt.

Why Hal9 agents?

Hal9 agents are intentionally simple: read from stdin with input(), write to stdout with print(). No framework lock-in. That shape maps cleanly to how MCPs expose tools to models, and makes agents easy to:

• Develop — plain Python, any library
• Deploy — hal9 deploy or GitHub Actions to hal9.com
• Interoperate — same agent can run as a chatbot, an API, or an MCP tool

Agents

send-email

Example prompt:

send email to javier@hal9.ai with text hello!

The agent:

1. Reads the prompt via input()
2. Calls a Groq model with a send_email tool definition
3. Maps the tool call to the Resend API
4. Prints the result via print()

Environment variables

Local run

cd send-email
pip install -r requirements.txt
export GROQ_API_KEY=...
export RESEND_API_KEY=...
echo "send email to you@example.com with text hello!" | python app.py

Deploy to Hal9

export HAL9_TOKEN=...   # from https://hal9.com/devs
hal9 deploy send-email --name send-email --access public \
  --title "Send Email" \
  --description "Send emails via Resend using natural language prompts"

On push to main, if files under send-email/ change, .github/workflows/send-email.yaml deploys a new version to Hal9 (same pattern as hal9ai/hal9 app deploy workflows). Set the HAL9_TOKEN repository secret in GitHub Actions.

Publish to the MCP Registry

The hosted send-email MCP is published as a remote-only server to the official MCP Registry under io.github.hal9ai/send-email. Metadata lives in send-email/server.json. On push to main (when send-email/ changes), .github/workflows/publish-send-email.yaml publishes it using GitHub OIDC — no extra GitHub secrets. The registry version is version in that file. If it is already published, the job skips. Bump version there to ship a new registry entry.

run-python

Example prompt:

install a package that calculates space orbits and compute the Hohmann transfer time from Earth to Mars and back, departing 2026-09-08

The agent:

1. Reads the prompt via input()
2. Calls a Groq model with install_packages and run_python tool definitions
3. Installs requested PyPI packages into a sandbox directory
4. Executes the generated Python in a subprocess and captures stdout/stderr
5. Prints the result via print()

Environment variables

Local run

cd run-python
pip install -r requirements.txt
export GROQ_API_KEY=...
echo "print the first 10 primes" | python app.py

Deploy to Hal9

export HAL9_TOKEN=...   # from https://hal9.com/devs
hal9 deploy run-python --name run-python --access public \
  --title "Run Python" \
  --description "Run Python code from natural language prompts, with optional PyPI package install"

On push to main, if files under run-python/ change, .github/workflows/run-python.yaml deploys a new version to Hal9. Set the HAL9_DEVEL_TOKEN / HAL9_NEXT_TOKEN repository secrets in GitHub Actions.

Publish to the MCP Registry

The hosted run-python MCP is published as a remote-only server to the official MCP Registry under io.github.hal9ai/run-python. Metadata lives in run-python/server.json. On push to main (when run-python/ changes), .github/workflows/publish-run-python.yaml publishes it using GitHub OIDC. Bump version in server.json to ship a new registry entry.

Website (GitHub Pages)

Static site lives in docs/ — no build step.

Enable Pages: repo Settings → Pages → Build and deployment → Source: Deploy from a branch → Branch: main → Folder: /docs.

Contributing an agent (MCP)

This repo is designed so coding agents (Claude Code, Grok Build, Cursor, Codex, …) and humans can contribute a new MCP with minimal guesswork. Full checklist: AGENTS.md. Use send-email/ as the reference implementation and .github/workflows/send-email.yaml as the deploy pattern.

Expected layout

# minimum
my-tool/
  app.py               # input() → work → print()
  requirements.txt     # optional
  hal9.yaml            # optional welcome

# also update
.github/workflows/my-tool.yaml
docs/agents.json
docs/my-tool/index.html   # dedicated docs page for the MCP
README.md

Minimal agent

# my-tool/app.py
prompt = input()
# … call APIs, tools, models …
print(result)

Steps

1. Create a new folder at the repo root, e.g. my-tool/. Keep the name short, kebab-case.
2. Add app.py. Use input() / print() (or any stdin/stdout) so the agent stays Hal9- and MCP-friendly. Prefer no hal9 package unless you need session state.
3. Add a requirements.txt if you need third-party packages.
4. Optionally add hal9.yaml with a welcome: message.

5. Add a GitHub Actions workflow, .github/workflows/my-tool.yaml, that deploys when that folder changes:

    on:
      push:
        branches: [main]
        paths:
          - my-tool/**
          - .github/workflows/my-tool.yaml
    # job: pip install hal9 → checkout → if my-tool/ changed:
    hal9 deploy my-tool --name my-tool --access public \
      --title "My Tool" --description "…"

   Secret: HAL9_TOKEN (agent runtime keys like GROQ_API_KEY are configured on the Hal9 side / local env — never committed).

6. Register the agent in docs/agents.json (include an id, description, and docs_path pointing at its docs page) and mention it in the table above.
7. Add a dedicated docs page at docs/<my-tool>/index.html so it shows up at https://hal9ai.github.io/mcp.build/my-tool/ (or https://mcp.build/my-tool/). Copy docs/send-email/index.html as a template — it covers what the MCP does, how agents use it, and how to add it to Claude and other MCP clients.
8. Do not commit secrets; document required env vars in this README.

License

Contributions are welcome. Individual agents may carry their own licenses; see each folder for details.