Let your AI ask before you publish.
Everything that gets published passes a person. That is the part that does not scale. Point whatever assistant you already use at our MCP server and it can check a draft itself, mid-conversation, before anyone spends a review on it.
https://api.dasho.ai/mcp
Four kinds of people use this
Marketing and brand teams
You approve what goes out and you carry it when it is wrong. This lets your assistant check a draft before it reaches you.
Agencies running several brands
Different rules per client, all in your head. Ask before you post rather than after a client flags it.
Regulated categories
Beauty, skincare, supplements, health, clinics. The categories where one caption costs more than the tool.
Developers and AI builders
You are wiring an assistant or an agent that drafts copy. Give it somewhere to check its own work.
Five things, weighted. Claims carry most of it.
score_content needs no key and no account. It scores a draft on five
dimensions and returns an overall number out of 100. The weights are fixed:
| Dimension | Weight | What it looks at |
|---|---|---|
| Regulated claims | 40% | Six families of wording a regulator acts on. It returns the exact phrase it found and why it is a problem, not a vague warning. |
| Readability | 15% | Sentence and paragraph length, and whether a reader gets through it. |
| Clear call to action | 15% | Whether the post actually asks for anything. |
| Brand voice | 15% | General voice quality: hype words, filler, machine cadence. |
| Content structure | 15% | Whether the piece is built well enough to read on a phone. |
score_vs_industry runs the same five and adds a comparison against a
benchmark for the industry you name.
Six families of wording, named one by one
It does not say "review this claim". It hands back the phrase it found, what family it belongs to, and why it is a problem.
Regulator approval
"FDA approved", "BFAD approved"
Clinical proof
"clinically proven", "scientifically proven"
Professional endorsement
"dermatologist approved", "doctor recommended"
Absolute safety
"100% safe", "no side effects"
Therapeutic claim
"cures acne", "treats eczema"
Guaranteed outcome
"guaranteed results", "permanent whitening"
Worked example. This caption goes in: "Clinically proven to erase wrinkles in 7 days. 100% safe, FDA approved, dermatologist approved." It comes back at 29 out of 100, with four claims named. The same caption without those four phrases comes back at 63. Call the endpoint and check.
Your own rules, which no free check can guess
The free check knows the law in general. It does not know what your brand promised. With a key, five more tools read your Brand Card: the say-this and avoid-this rules your team wrote down, your voice, and your audience.
check_compliance | Scores a draft against your say-this and avoid-this rules, and names each rule it broke. |
get_brand_context | Returns your brand voice, rules and target audience, so an assistant can write to them in the first place. |
rewrite_content | Rewrites a post so it passes your own rules. |
generate_report | Pulls a governance report for the brand. |
get_governance_trend | How the brand has scored week over week. |
These need X-Dasho-Api-Key and a brand on your workspace. That is the
product. The free tools cannot see any of it, by design.
One server URL. That is the setup.
{
"mcpServers": {
"dasho": { "url": "https://api.dasho.ai/mcp" }
}
}
Plain JSON-RPC 2.0 over HTTP. Call tools/list to see what is available,
tools/call to run one. Keyed tools take an X-Dasho-Api-Key header.
Not editing a config file by hand? Step-by-step for ChatGPT and Claude, including what to do when your assistant says it cannot install it.
The things people ask first
Do I need an account?
Not for the two free tools. No key, no signup, no card. The tools that read your own brand rules need a key, because they need a brand to read.
What does it cost?
The free tools are free and stay free. They run on our own servers with no model call behind them, so they cost us close to nothing and we have no reason to meter them.
Is my content stored?
Content sent to the free tools is scored and returned. It is not added to a brand, not used to train anything, and not visible to our team.
What is MCP?
Model Context Protocol. An open standard that lets an AI assistant call an outside service. It means your assistant can ask us mid-conversation instead of you leaving to check a dashboard.
Your rules are the part worth paying for.
Checking against general good practice is free and always will be. Checking against what your brand actually promised needs a Brand Card, and that is the product.