How to connect brand monitoring to Hermes Agent

You can spend $200 a month on a SaaS brand-monitoring tool. Or you can spend ten minutes wiring brand monitoring directly into Hermes Agent and pay two cents per check.

I picked the second option. Here is how it went.

Hermes Agent shipped MCP OAuth 2.1 support in its April 8 release. That detail matters. It means any API that exposes itself as an MCP server can be plugged into Hermes in three lines of YAML. MentionsAPI exposes itself as one. So tracking your brand across ChatGPT, Claude, Gemini, and Perplexity through Hermes is no longer a "someday" project. It is a one-evening task.

This article walks through every step. Setup, first call, Slack alerts, real cost numbers, and the two mistakes I keep seeing people make.

Why brand monitoring belongs inside Hermes Agent

Most brand-monitoring tools are SaaS dashboards. You log in, you look at charts, you log out. The tool lives somewhere else.

Hermes is different. It lives on your server. It has persistent memory. It can talk to you on Slack, Discord, or any of the seventeen messaging platforms it supports out of the box.

Most brand-monitoring SaaS dashboards force you to log in and scan for changes. Hermes inverts that. You only hear from it when there is something to hear.On daily monitoring loops

That last bit is the one that sells me. When ChatGPT stops mentioning my brand, I do not want to log into a dashboard to find out. I want a Slack DM at 9:02 AM telling me what changed and which competitor took the slot.

There is also a memory argument. Hermes remembers what it learned yesterday. So when it pings you about a brand mention, it can also tell you "this is the third time this week and the second competitor that has shown up in the same answer." That kind of pattern detection requires state. Most brand-monitoring SaaS does not actually keep that state in a way you can query. Hermes does.

What you need before you start

Three things.

First, Hermes Agent installed on a server you control. A small VPS, your laptop, a Raspberry Pi, anything that runs Linux, macOS, or WSL2. The official install command is one line:

curl -fsSL https://raw.githubusercontent.com/NousResearch/hermes-agent/main/scripts/install.sh | bash

Full setup instructions live at the official Hermes quickstart.

Second, a MentionsAPI account with at least $10 in your wallet. MentionsAPI is pay-as-you-go: top up $10, every API call deducts from your balance, no subscription. A cache hit is $0.02. A full multi-provider fan-out with web search is $0.75. We will mostly use the cheaper modes.

Third, the brand or brands you want to track. Yours, your competitors, or both.

Setting up the MentionsAPI MCP connection

Hermes reads its configuration from ~/.hermes/config.yaml. If you used the install script, this file already exists. Open it.

Look for the mcp_servers: block. If it is not there, add it at the top level. Then drop in this:

mcp_servers:
  mentionsapi:
    url: "https://mentionsapi.com/mcp/sse"
    headers:
      Authorization: "Bearer YOUR_MENTIONSAPI_KEY"
    connect_timeout: 60
    timeout: 180

Replace YOUR_MENTIONSAPI_KEY with the actual key from your MentionsAPI dashboard.

Save the file. In Hermes, type /reload-mcp. You should see a confirmation that mentionsapi connected and three or four tools were registered (check_brand, discover_queries, extract_brands, and a usage tool).

This is the entire integration. No SDK. No middleman service. Hermes is now wired directly to MentionsAPI through MCP.

Your first brand-monitoring call

Open Hermes. Type something like this:

Use the mentionsapi tool to check whether ChatGPT mentions Linear when I ask "best project management tools for engineers." Show me where Linear ranked and what it was compared to.

Hermes will figure out which tool to call, format the request, send it, and parse the response. You should see something like:

That answer took about 1.6 seconds end-to-end and cost $0.02 because it hit the cache. Subsequent runs of the same prompt over the next 24 hours will also hit the cache. If you want a fresh result, add cache_bypass: true and pay $0.10 to $0.75 depending on the mode.

Now ask Hermes a more useful question:

Run that same check for Linear, Asana, Jira, ClickUp, and Notion. Tell me each one's rank, sentiment, and citation URL.

This uses the same MCP tool but loops across five brands. Hermes will call the tool five times and aggregate the results. Total cost for this five-brand sweep: about ten cents.

Sending Slack alerts when something changes

The point of monitoring is to know when something changes, not to look at a static report.

Here is how to wire that up. In Hermes, define a recurring task. The exact syntax depends on your version, but it looks like this:

Schedule a daily brand check at 09:00 UTC.

Run mentionsapi.check_brand for "Linear" against the prompts:
- best project management tools for engineers
- alternatives to Jira
- linear vs asana

For each result, compare to yesterday's result (stored in your memory).
If Linear's rank dropped, citation count dropped, or a new competitor appeared,
post a Slack message to #brand-watch with what changed.
If nothing changed, do nothing.

Hermes will:

1. Run the three prompts every morning at 9 AM UTC.
2. Compare today's results to yesterday's (using its persistent memory).
3. Post a Slack message only when something is worth your attention.

This is the killer feature. Most brand-monitoring SaaS dashboards force you to log in and scan for changes. Hermes inverts that. You only hear from it when there is something to hear.

I have been running this exact loop for two weeks. I get one or two Slack messages a week. Both have been useful. One was a competitor swap. One was a citation source change. Neither was noise.

What this costs you per month

Specific numbers. Three prompts, one brand, twice a day, every day.

The math:

• 108 cached calls × $0.02 = $2.16
• 72 fresh calls × $0.18 = $12.96
• Total: $15.12 per month

Compare that to the AEO and brand-monitoring SaaS landscape. Otterly starts at $49 a month for a single brand. Frase is bundled at $45 a month minimum. Profound is enterprise-priced ("contact us"). The seven best AEO/GEO tools for 2026 reviewed by Scrunch all cluster in the $50 to $200 range for a single brand.

The cost story is the picks-and-shovels argument in numbers.

Adding competitor tracking to the same loop

Once basic monitoring is running, the next step is adding competitors. This is where the cost-per-value ratio gets ridiculous.

Tell Hermes to track three competitor brands alongside yours. Pass them as a single array to MentionsAPI's track_brands parameter. The API runs one extraction pass and returns positions for all four brands in one response. You pay for one call instead of four.

Now the daily comparison gets richer. Hermes can tell you "your rank held steady, but Competitor B jumped from position 5 to position 2 on the 'best AI brand monitoring' prompt." That is a direct signal that someone else shipped content or won a backlink that AI engines now treat as authoritative.

Five brands tracked daily, one of yours plus four competitors, three prompts: about $25 per month. That is a research assistant for the price of one paid Slack seat.

The mistakes I keep seeing people make

Two patterns. Both kill the value.

Mistake one: people add too many prompts on day one. They start with twenty queries because they want full coverage. The result is alert fatigue. The Slack channel becomes noise. They mute it within a week.

The fix is to start with three prompts. Ones you genuinely care about. The exact phrasing your customers use, not the broad category you hope to rank for. After two weeks, add more if you have headroom. Never start with twenty.

Mistake two: people forget that Hermes has memory. They configure brand monitoring as if each run is independent. So Hermes ends up sending alerts every day even when nothing changed. The trick is in the prompt to Hermes: tell it explicitly to compare today's result to yesterday's result and only alert on deltas.

You can get even smarter. Tell Hermes to only alert if the rank dropped by two or more positions, or if the citation source changed. That kind of stateful filtering is exactly what Hermes is designed for. Most SaaS tools cannot do this because they do not have a memory layer that you can program against.

Frequently asked questions

Wire it up tonight

The whole setup is ten minutes if you already have Hermes installed. Add the MCP block to your config, reload, and run your first check. Then write the daily-monitoring prompt and forget about it.

Start with three prompts and one brand. Let it run for a week. Look at what Slack messages you actually got and whether they were useful. Adjust from there.

The whole point of this stack is that it stays out of your way until something is actually worth your attention. That is the test for whether your monitoring is set up right. If you log into a dashboard daily, your monitoring is wrong. If Slack pings you when something changes, your monitoring is right.