We Don't Care For Dashboards

beehiiv MCP, GRADE: A

Growth & Community Building

beehiiv

Description: Beehiiv is a newsletter platform that now acts more like an operating system for an audience. You can write and design the newsletter, manage the site and domain, add images, run automations, and pull performance data back into the rest of your business. Its MCP lets an AI agent work directly inside the publication, which is the direction products are going.

Use case: You are a business owner or creator with something useful to teach, but you don’t want a newsletter to become another full-time job. Beehiiv gives you one place to publish, grow, segment, monetize, and understand the audience. The MCP becomes useful once AI is already part of your workflow and your agent can manage the repetitive work.

Brody’s Review: I had not used Beehiiv in about a year but I came back to build Smoke Test and expected the same newsletter product I remembered…. I was shocked. It is a much more robust product now.

The custom-domain setup was seamless, and the MCP gives us 146 tools (!) covering posts, design, media, subscribers, segments, automations, analytics, referrals, surveys, paywalls, web pages, sponsorships, and podcasts. We have not fully cracked into most of that, so I am not going to pretend we tried everything.

What we did test was the important part: getting the publication live, connecting it to our Hermes Agent stack, and making the content and metrics available to the rest of our systems. Flawless.

Crustdata, GRADE: A-

Outbound & Data Enrichment

Crustdata

Description: A real-time B2B data API built for agents. People + company search, enrichment, and live signals (job changes, funding, hiring spikes, web-traffic jumps) pulled from 16+ sources and kept fresh by live crawl. The tool has a native MCP, so agents can interact with it directly. Pricing is credit-based, searches run ~0.03 credits each, enrichment ~1 credit per person, and self-serve pay-as-you-go runs ~3x the price ($0.20 - $0.30 per credit).

Use case: You want to run outbound & enrichment at real volume, or building an agent that needs live people + company data on tap. You can ask for “tech companies in Austin that raised $10M in 2025,” get a clean list, enrich it, set a watcher so you're pinged the moment someone switches jobs or raises. Signals on tap, which is wicked.

Jake’s Review: We ran our own enrichment pass and clocked ~90% accuracy, comfortably above the usual off-the-shelf stack, which lands closer to 60-70% in our internal testing.

Real-time is the main differentiator. Incumbents (ZoomInfo, PDL, Apollo) ship batch data on a monthly refresh, but Crustdata live-crawls, so a headline changed an hour ago shows up in seconds, and search runs across all of LinkedIn, not just a stored database.

Watching every LinkedIn post that matches a keyword + persona is a sharp way to find who's talking about your thing right now. The data starts to thin out for contact details, null lookups waterfall to third parties, and coverage skews hard US (APAC is patchy, but that’s ok since we aren’t targeting that market). Expensive, but it earns its keep when volume or an agent product justifies it.

Instantly, GRADE: B

Outbound Infrastructure & Sequencing

Instantly · Official campaign analytics

Description: Instantly is outbound infrastructure for connecting inboxes and domains, warming them, enrolling contacts, sequencing messages, and reading campaign performance. It covers the operational path from “we have a list” to “messages are going out without nuking the domain,” with API access for teams that want the data somewhere else.

Use case: You run repeatable outbound and need one place to manage sender health, enrollment, sequences, and reporting. The warmup and sending guardrails matter when a campaign must scale without making every inbox look like a machine. The API matters when the platform is one part of a larger operating system rather than the operating system itself.

Brody’s Review: We have used Instantly across the entire outbound workflow: inbox and domain setup, warmup, contact enrollment, sequencing, copy changes, sending, analytics, and reporting. The automated warmup works really well, sequencing is adequate, and with enough customization and guardrails to make sending behave more organically and protect the domain. The analytics are good, and the API makes it straightforward to pull metrics into our internal and client Hermes Agent deployments.

The front end kinda sucks because it constantly times out, campaign setup and setting config are harder than they should be, and basic jobs like adding lists or editing copy take too much effort. We have had better luck operating Instantly headlessly through Hermes Agent, which is where products are heading anyway, so I can’t hate on the dashboard too much since I don’t use it to drive my outbound.

The grade stays at B because the infrastructure works, but we are starting to test cheaper, simpler replacements.

Hermes Agent, GRADE: A+

Agent Harness & Orchestration

Nous Research · Hermes Agent

Description: Hermes Agent is an open-source agent harness from Nous Research. It gives agents persistent memory, reusable skills, tools, recurring schedules, and isolated subagents. Operators can run it across machines and messaging platforms while switching models and providers without rebuilding the system around each one.

Use case: You want an agent to own recurring work instead of answering one prompt at a time. Hermes Agent fits workflows that must remember decisions, reuse procedures, divide work across agents, connect to external tools, and report back where the operator already works. It’s a l

Brody’s Review: Where do I start without glazing too much? The product is awesome, the team is fantastic, and their vision aligns with ours.

Hermes Agent is our go-to harness across the biz and client systems. Persistent memory carries context across sessions, which helpful for small model context window compactions, skills turn successful procedures into reusable operating knowledge, and the agents can split independent workstreams into subagent tasks without stuffing everything into one context window. The looping and scheduling mechanisms have worked flawlessly for the recurring, self-healing tasks we run.

In my setup I pair it with Nous Portal for model access (OpenRouter as my fallback), which lets me easily benchmark and switch between different models. That matters because the best model changes by job and because we would rather own the workflow than have it trapped inside one model vendor.

We still have not tapped the full breadth of capabilities because the team ships faster than we can test them, which is the rare kind of limitation that strengthens the A+.

TulexAI, GRADE: C+

Model Aggregation & Routing

TulexAI · Official product artwork

Description: Get access 30+ frontier models with one subscription. Claude, GPT, Gemini plus image, video, voice, etc etc all drawn from a single monthly “platform token” budget. $11-$99/mo with no card to start. They claim it's ~70% cheaper than stacking ChatGPT + Claude + Midjourney + the rest of the pile.

Use case: You want to sandbox a pile of models, or lean on a cheap auto-router, without paying per token or juggling five subscriptions. Good for non-sensitive work like drafting, throwaway image gen, and comparing how three models answer the same prompt. It’s bad for sensitive data and their TOS implies your data is routed to the larger models for them to monetize or use for product development…. which is pretty standard from the big labs anyways.

Jake’s Review: It's close to, if not, the cheapest way to touch every major model behind one bill. The catch is what makes it that cheap, and who's holding it.

On data: the privacy page says the right words (“doesn't sell your info,” “doesn't train on your prompts”) but it routes every prompt out to the underlying providers and explicitly punts on what they do with it. The translation here is no zero-retention guarantee, you don't pick the route, and the model providers can do whatever they want with that data.

It's a small shop (Latvia contact, hosted on Railway + Supabase) with essentially no independent review trail, its own yellow flag in a category crawling with sketchy resellers. Our read on the 70%-cheaper math is boring gym-membership economics: most subscribers never burn their token cap. Where it thins out: don't pipe client data or anything sensitive through it. If you need routing with privacy controls, OpenRouter lets you switch logging off and restrict to providers that won't train on your data and of course, Nous Portal. The trade-off is both are pay-per-use.