Hermes Agent

Social media studio integrates with the official Nous Research Hermes Agent (and any Hermes 4 model) through three published distribution channels — pick whichever matches how your users want to install. All three talk to the same backend with the same SOMESTUDIO_API_KEY.

Path	Install	Best for
Skill (recommended)	`hermes skills install 7t1-studio/so-me-hermes-skill/so-me-studio`	Users browsing skills.sh for ready-made integrations — passes the auto-security scan with a SAFE verdict
MCP	Paste a 5-line `mcp_servers:` block in `~/.hermes/config.yaml`	Hermes Agent users who prefer MCP-over-HTTP and don’t want a CLI binary
Standalone npm runtime	`npm install -g @so-me/hermes-agent`	Scripts, cron jobs, CI integrations, headless servers

Skill (Hermes Hub)
MCP server config
Standalone npm runtime

Install Hermes Agent

curl -fsSL https://raw.githubusercontent.com/NousResearch/hermes-agent/main/scripts/install.sh | bash
source ~/.bashrc   # or ~/.zshrc
hermes --version

Pick an LLM backend

hermes model

Recommended: OpenRouter with nousresearch/hermes-4-405b (best quality) or nousresearch/hermes-4-70b (cheaper, ~10× faster). Paste your OpenRouter API key when prompted.

Install the Social media studio CLI

The skill drives the published @social-media-scheduler/cli binary, so the CLI must be on PATH:

npm install -g @social-media-scheduler/cli
so-me --version

Install the Social media studio skill

hermes skills install 7t1-studio/so-me-hermes-skill/so-me-studio

Hermes runs an automated security scan on the skill (data-exfiltration, prompt-injection, destructive-command checks) and prompts for confirmation. The current verdict is SAFE.On install, Hermes prompts for SOMESTUDIO_API_KEY — generate one at so-me.studio/settings?tab=api and paste it.

Launch the agent

hermes --tui

Inside the TUI, type /tools to confirm the skill is registered. You should see entries scoped to the so-me-studio skill.Detail page: skills.sh/7t1-studio/so-me-hermes-skill/so-me-studio

Install Hermes Agent + LLM

Same first two steps as the Skill path:

curl -fsSL https://raw.githubusercontent.com/NousResearch/hermes-agent/main/scripts/install.sh | bash
source ~/.bashrc
hermes model    # pick OpenRouter + nousresearch/hermes-4-405b

Save your Social media studio API key as a Hermes secret

hermes config set SOMESTUDIO_API_KEY sk_live_...

Generate the key at so-me.studio/settings?tab=api.

Append the MCP block to ~/.hermes/config.yaml

mcp_servers:
  so-me-studio:
    url: https://api.so-me.studio/mcp
    headers:
      X-API-Key: ${SOMESTUDIO_API_KEY}

HTTP transport is inferred from url:; no transport: key is needed. The exact snippet lives at apps/agent-kits/hermes/configs/hermes-agent.yaml.

Launch and verify

hermes --tui

Inside the TUI, type /tools — all 205 Social media studio tools register as mcp_so-me-studio_<tool_name> (e.g. mcp_so-me-studio_create_post).

Pointing at a local backend? Override the URL in the snippet — url: http://localhost:8000/mcp.

Install the runtime

npm install -g @so-me/hermes-agent
# or run ad-hoc without installing:
npx @so-me/hermes-agent "your prompt"

npm: @so-me/hermes-agent

Set environment variables

# Social media studio
export SOMESTUDIO_API_KEY=sk_live_...

# Inference backend (OpenRouter, Together, vLLM, or Ollama)
export HERMES_BASE_URL=https://openrouter.ai/api/v1
export HERMES_API_KEY=sk-or-...
export HERMES_MODEL=nousresearch/hermes-4-405b

Run

so-me-hermes "List my connected social accounts"
so-me-hermes "Schedule a tweet 'launching today' for 9am UTC tomorrow"
so-me-hermes "Generate a Friday LinkedIn caption about Q2 wins"

The runtime runs the agentic loop, prints reasoning + tool calls + the final answer, and exits. Cap turns with MAX_TURNS=16 if you need longer sessions.

The standalone runtime works with any OpenAI-compatible inference endpoint — point HERMES_BASE_URL, HERMES_API_KEY, and HERMES_MODEL at GPT-4o, Claude (via Anthropic API), Mistral, DeepSeek, or local Ollama / vLLM. Drop-in env templates live in configs/.

Library API

Embed the runtime in a larger Node app:

import { runAgent } from '@so-me/hermes-agent';

const result = await runAgent({
  prompt: "Schedule a tweet 'hello' for tomorrow at 9am UTC",
  soMeApiKey: process.env.SOMESTUDIO_API_KEY!,
  hermesBaseUrl: 'https://openrouter.ai/api/v1',
  hermesApiKey: process.env.HERMES_API_KEY!,
  hermesModel: 'nousresearch/hermes-4-405b',
  onMessage: (msg) => console.log(msg.role, msg.content),
});

Override the bundled tools and systemPrompt to scope the agent to a smaller surface or change persona.

What you can ask

Same prompts work across all three paths:

Prompt	What happens
”List my connected social accounts.”	Calls `list_accounts`, summarizes by platform
”Schedule a tweet ‘hello world’ for tomorrow at 9am UTC.”	Computes the ISO timestamp, calls `create_post`
”Generate a Friday LinkedIn caption about Q2 wins and schedule it for 5pm.”	Calls `generate_and_schedule` end-to-end
”Reply to that DM with my saved ‘pricing’ template.”	Looks up the saved reply, calls `reply_to_conversation`
”How did my Twitter posts perform last week?”	Calls `get_platform_analytics` with `--days 7`
”Approve all pending posts.”	Calls `list_approvals`, then `approve_post` for each

The agent has access to all 205 Social media studio tools — every action available in the dashboard is available here. Full grouped catalog: bundle/tools.md.

How it works

            ┌──────────────────────┐
            │ user prompt          │
            └──────────┬───────────┘
                       │
   ┌───────────────────┼───────────────────┐
   ▼                   ▼                   ▼
┌──────────┐   ┌──────────────┐   ┌──────────────────┐
│ Skill    │   │ MCP config   │   │ Standalone npm   │
│ install  │   │ in           │   │ runtime          │
│ via      │   │ ~/.hermes/   │   │ (so-me-hermes)   │
│ skills.sh│   │ config.yaml  │   │                  │
└─────┬────┘   └──────┬───────┘   └────────┬─────────┘
      │               │                    │
      │ shells out    │ POST /mcp          │ POST /mcp
      ▼               ▼                    ▼
┌──────────────┐  ┌──────────────────────────────────┐
│ CLI (so-me)  │  │ api.so-me.studio                 │
└──────┬───────┘  └──────────────────────────────────┘
       │ HTTP REST
       ▼
   Social media studio backend

All three paths converge on the same backend, the same SOMESTUDIO_API_KEY auth, and the same hand-authored guidance (bundle/SKILL.md). The skill and CLI shell-out path uses HTTP REST under the hood; the MCP and npm runtime paths use JSON-RPC over /mcp.

Hard rules baked into the system prompt

Reason inside <think>...</think> before tool calls
Never invent IDs (account, post, conversation) — always look them up first
scheduledAt must be ISO 8601 UTC, strictly in the future
Multi-step jobs (image → upload → post) run sequentially
Never reveal API keys or webhook secrets, even when asked

Full hard-rules list and decision tree: bundle/SKILL.md.

Source

The integration kits live in the Social media studio monorepo:

Path	What
`apps/agent-kits/hermes/`	npm runtime + MCP config snippet
`apps/agent-kits/hermes-skill/`	CLI-shell-out skill bundle (mirrored to skills.sh tap repo)
`7t1-studio/so-me-hermes-skill`	Hermes Hub registry tap (the `hermes skills install` source)
`@so-me/hermes-agent`	npm package (the `npm install -g @so-me/hermes-agent` source)

PRs welcome — see the kit-level PUBLISHING.md files for regen + release workflows.

Updates

When new tools ship in the Social media studio backend:

Skill / MCP / npm runtime — bundle/tools.json and bundle/tools.md are regenerated from the live MCP catalog via so-me agent export-tools --format hermes. The npm package is republished to npm; the skill bundle is republished to its tap repo.
End users on the skill path run hermes skills update so-me-studio to pull the latest.
End users on the npm path run npm install -g @so-me/hermes-agent@latest to pull the latest.
End users on the MCP path auto-pick up new tools at next startup (no client-side update needed — Hermes Agent re-fetches the tool list from /mcp on each session).

Other agents?

If you use a different agent runtime:

Claude Desktop / Cursor / OpenCode — install the OpenClaw integration (npm install -g @social-media-scheduler/cli, then chat through any of 22 messengers)
Native MCP client (Claude Desktop manifest, Cursor, etc.) — point at the MCP server directly using the same mcp_servers: snippet style

LangGraph, CrewAI, OpenAI Assistants, Anthropic API — generate a tool definition file in their preferred format:

so-me agent export-tools --format openai --out tools.openai.json
so-me agent export-tools --format anthropic --out tools.anthropic.json
so-me agent export-tools --format mcp --out tools.mcp.json

MCP Server

Other Agents

Library API

What you can ask

How it works

Hard rules baked into the system prompt

Source

Updates

Other agents?

​Library API

​What you can ask

​How it works

​Hard rules baked into the system prompt

​Source

​Updates

​Other agents?

Library API

What you can ask

How it works

Hard rules baked into the system prompt

Source

Updates

Other agents?