Don't have an agent yet?
Start here.

Agent Media Tools gives AI agents hands — images, PDFs, STT, pastes, webhooks. First pick an agent that can call tools. Use the left menu to jump to any section.

Hermes Claude Desktop Local / Ollama ChatGPT MCP + REST
Mira — get started with agents
We don’t host your agent — we host the tools it calls. Official product docs win on edge cases; we focus on “get something working today.” Costs are approximate (checked ~July 2026) and change — verify with each provider.

1 · Pick your path

Click a card to jump to the full setup for that stack.

2 · Rough cost comparison

Hobbyist light daily use — not a quote.

Stack What you run Typical monthly $ MCP? Best for
Hermes + local Hermes + Ollama ~$0 API (+ power/GPU) Privacy, experiments
Hermes + API OpenRouter / DeepSeek / etc. Often $5–40 if careful Capable without big GPU
Claude Desktop Anthropic app + MCP Claude Pro/Max or API ✅ native Fastest tools-in-chat
Open WebUI + Ollama Browser UI + local models ~$0 API OpenAPI / tools Local chat UX
ChatGPT Custom GPT + Actions Plus / Team REST Actions Light workflows
Agent Media Tools Hosted tools (any agent) Free tier + credits / $5 Premium ✅ MCP + REST Media, PDF, STT, glue

Always check live pricing for Anthropic, OpenAI, OpenRouter, DeepSeek, xAI, etc. Our pricing →

3 · Hermes Agent — full setup

🧠 Recommended for “real” agents

Goal: Hermes installed → model talking → Agent Media Tools MCP loaded → first tool call succeeds.

1

Install Hermes

Use the official installer for your OS. Desktop is easiest on Mac/Windows; CLI is fine on Linux/VPS.

Docs: hermes-agent.nousresearch.com/docs

# Linux / macOS / WSL — confirm current command on their site
curl -fsSL https://hermes-agent.nousresearch.com/install.sh | bash

# Then open a new terminal and check
hermes --help
Terminal — after install
$ hermes --help
Hermes Agent — commands: chat, setup, tools, …
# If "command not found", re-open the terminal or check PATH

You should see Hermes CLI help — not “command not found.”

2

Connect a model (brain)

Hermes needs an LLM. Pick one:

  • Cheapest / private: Ollama local (see Local section)
  • Cheap API: OpenRouter or DeepSeek (OpenAI-compatible base URL + key)
  • Strong hosted: Anthropic / OpenAI / Nous Portal / etc.
# Example: guided setup (flags may vary — use hermes setup --help)
hermes setup

# Or set provider env vars in ~/.hermes/.env as their docs describe
Tip: Start with a small/cheap model. You can upgrade later. Tool-calling models work better than pure chat models.
3

Add Agent Media Tools as MCP

Option A — download our Hermes pack and follow the README inside:

agent-media-tools-hermes.tar.gz

Option B — point Hermes MCP config at a local clone or hosted URL. Typical pattern (paths vary by install):

# In Hermes MCP / config (YAML-style example)
mcp_servers:
  media-tool:
    command: "node"
    args: ["/absolute/path/to/mcp-server.js"]
    env:
      MEDIA_TOOL_URL: "https://agentmediatools.com"
      # optional for paid/agent endpoints:
      # MEDIA_TOOL_AGENT_KEY: "mt_your_key_here"
What “success” looks like
MCP servermedia-tool connected
Toolscreate_paste, screenshot_page, resize_image, generate_image, …
Test“Create a paste saying hello via media tools”

In Hermes tool list / MCP UI you should see Agent Media Tools functions after reload.

More: Agent Hub · AI agents guide

4

Smoke test

In chat with Hermes:

  • “Use Agent Media Tools to create a paste with the text: hello hermes”
  • “Screenshot https://example.com and describe what you see” (free daily quota applies)
If tools never appear: restart Hermes, check absolute path to mcp-server.js, Node ≥ 18, and that MEDIA_TOOL_URL is reachable.

4 · Claude Desktop — full setup

🟣 Easiest path to MCP tools

Goal: Claude Desktop open → extension installed → tools usable in a normal chat.

1

Install Claude Desktop

Download from claude.ai/download, sign in, finish any onboarding Anthropic requires.

Claude Desktop (illustrative)
AppClaude · Desktop · signed in
You needA plan that allows Desktop + MCP/extensions (check Anthropic’s current plans)
2

Install our MCP extension (1 click)

Download agent-media-tools.mcpb

Double-click the file (or open it with Claude Desktop). Approve install when prompted. Fully quit and reopen Claude if tools don’t show.

After install
ExtensionsAgent Media Tools enabled
Toolscreate_paste · resize_image · screenshot_page · generate_image · …
Try saying“Generate a QR code for https://agentmediatools.com”

Exact Settings UI labels change — look for MCP / Extensions / Connectors.

Deep dive: Claude MCP setup (5 min)

3

Fallback: custom instructions (no MCP install)

If you can’t install extensions, paste our instructions instead:

Claude custom instructions guide (copy-paste block)

Tip: MCP is better — Claude calls tools natively. Custom instructions only teach it our REST API.
4

Optional API key for premium / credits

Free tools work without a key. For image gen, higher limits, or agent-only routes: create a free account on agentmediatools.com → Account → API Keys → create key → put it where Claude/MCP env expects MEDIA_TOOL_AGENT_KEY if your pack supports it.

5 · Local Ollama (+ Open WebUI) — full setup

💻 Private & low API cost

Goal: local model answers → (optional) chat UI → tools via Hermes or Open WebUI OpenAPI.

1

Install Ollama + pull a model

Install from ollama.com, then pull a model that supports tool use when possible:

# Example — model names change; check ollama.com/library
ollama pull llama3.2
ollama run llama3.2
Terminal
$ ollama run llama3.2
>>>
send a message… model replies locally
2

Choose how you chat / act

  • Agent mode: Point Hermes at Ollama’s OpenAI-compatible endpoint (usually http://127.0.0.1:11434/v1)
  • Chat UI: Install Open WebUI, connect Ollama, then add our OpenAPI tools

Open WebUI walkthrough: Open WebUI integration

3

Wire Agent Media Tools

With Hermes: same MCP steps as Hermes §3. With Open WebUI: import https://agentmediatools.com/openapi.json (or a trimmed subset) as tools/actions.

Reality check: Small local models are weaker at complex multi-tool plans. Start with paste / QR / simple resize before pipelines.

6 · ChatGPT Custom GPT — full setup

🟢 Lightweight Actions

Goal: Custom GPT can call our REST API for a few tools. Not 24/7 autonomy.

1

Create a Custom GPT

In ChatGPT (Plus/Team as required): Explore GPTs → Create → Configure. Name it e.g. “Media Tools Helper”.

2

Add Actions from OpenAPI

Import schema: https://agentmediatools.com/openapi.json

If the editor rejects a huge schema, import only the paths you need (paste, QR, resize, etc.) — see our guide.

Actions (illustrative)
SchemaImport from URL → agentmediatools.com/openapi.json
AuthNone for free tools · API key header for premium routes
TestTest action

Full guide: ChatGPT Custom GPT

3

Test in chat

“Create a short paste that says hello from ChatGPT using your action.”

Custom GPTs won’t match Hermes for cron, multi-hour jobs, or messaging gateways. Graduate to Hermes/Claude when you outgrow Actions.

7 · Connect Agent Media Tools (any agent)

Once something can call HTTP or MCP:

  • Discovery files: /llms.txt · /capabilities.json · /openapi.json · /mcp.json
  • Free tools: pastes, UUID, many image/PDF ops (rate limits) — no key
  • API key: site Account → keys · or POST /api/agent/provision
  • MCP packs: Claude .mcpb · Hermes tarball · remote MCP + Bearer
  • Credits: STT / remove-bg / screenshot / HTML→PDF / image gen after free daily quotas — Pricing

Terminal smoke tests

# 1) No auth — create a paste
curl -sS -X POST https://agentmediatools.com/api/paste \
  -H 'Content-Type: application/json' \
  -d '{"content":"hello from my new agent","title":"first paste"}'

# 2) With agent key — see credits + paid tool quotas
curl -sS https://agentmediatools.com/api/agent/status \
  -H 'Authorization: Bearer mt_YOUR_KEY'

# 3) Buy credits as an agent (returns Stripe URL or charges if PM saved)
curl -sS -X POST https://agentmediatools.com/api/agent/shop/checkout \
  -H 'Authorization: Bearer mt_YOUR_KEY' \
  -H 'Content-Type: application/json' \
  -d '{"item":"credits-100"}'
Example paste response
{
  "success": true,
  "url": "https://agentmediatools.com/p/…",
  "raw_url": "https://agentmediatools.com/p/…/raw"
}

If you get a URL back, your machine can reach us — the agent can too.

8 · First-win checklist

Tick these mentally before you dig into fancy pipelines:

  • Agent app runs and answers a normal question
  • MCP or Actions lists Agent Media Tools (or curl paste works)
  • Created a paste or QR code successfully
  • (Optional) Free account + API key saved for premium routes
  • (Optional) Know where credits live: Account → Credits · /pricing

You’re one paste away

Agent running? Wire tools. Still choosing? Hermes for power, Claude for easiest MCP.

9 · FAQ

Do I need Premium to try?
No. Free tools and free daily quotas work without a paid plan.

Hermes vs Claude?
Hermes: always-on, messaging, cron, full agent lifestyle. Claude Desktop: polished chat + excellent MCP for interactive sessions.

Grok / DeepSeek / OpenRouter?
Use them as model backends for Hermes (or similar). We host the media/tool API, not those models.

Tools don’t show up in MCP
Restart the host app, check absolute paths, Node 18+, and that the MCP process isn’t erroring on start. Try the curl paste test to isolate network issues.

Full tool list?
/tools · /llms.txt · /docs