Mentio

Mentio

Mentio watches developer platforms for the words a company cares about (its product, its competitors, its space), scores every post that contains them for relevance, sentiment and intent, and exposes the result as a REST API. This skill is that API: read the rule file for an area before the first call to it, then call it with curl or the mentio CLI.

• Base URL: https://api.mentio.dev/v1
• Auth: Authorization: Bearer $MENTIO_API_KEY on every request
• Docs: https://docs.mentio.dev (every page is also served as Markdown at https://docs.mentio.dev/<page>.mdx)

Setup

The key lives in the MENTIO_API_KEY environment variable. When it is unset, or a request answers 401, stop and tell the user how to fix it:

1. Create a key at https://app.mentio.dev/api-keys. A write key runs the whole skill; a read key covers searching, analytics and people but no changes.
2. echo 'MENTIO_API_KEY=mk_live_...' >> ~/.openclaw/.env
3. openclaw gateway restart

Never print the key back, write it into a file the user did not ask for, or send it anywhere but api.mentio.dev. If you are unsure the key works, GET /v1/keywords is a cheap check: 401 means the key is wrong, an empty list means nothing is tracked yet.

How to call the API

When the mentio CLI is on the PATH (command -v mentio), prefer it: it reads MENTIO_API_KEY, prints JSON, and names every endpoint noun:verb (rules/cli.md). Otherwise use curl. Reads put the filters in the query string; writes send JSON.

# Read
curl -sS "https://api.mentio.dev/v1/mentions?relevant=true&sentiment=negative&limit=20" \
  -H "Authorization: Bearer $MENTIO_API_KEY"

# Write
curl -sS -X POST "https://api.mentio.dev/v1/keywords" \
  -H "Authorization: Bearer $MENTIO_API_KEY" -H "Content-Type: application/json" \
  -d '{"term": "acme", "kind": "brand"}'

Pipe through jq when it is available to pick fields (| jq '.data[] | {url: .post.url, sentiment: .classification.sentiment}').

The API in ten rules

1. Ids are prefixed strings: kw_ keywords, mm_ mentions, aut_ people, seg_ segments, feed_ alerts, dest_ channels, key_ API keys. An id from another workspace is a 404, never a 403.
2. Lists come back as { data, nextCursor } (pass nextCursor back as cursor with the same filters and sort; null on the last page) or, for people, { data, total } with limit and offset. Configuration lists (keywords, alerts, channels, segments, API keys) are complete in data. Single resources come back bare.
3. One write per resource: PATCH with only the fields to change, null clears a field. POST on a collection creates and answers 201 with the resource; DELETE answers 204; actions are POST verbs on the resource (/people/{id}/merge, /alerts/{id}/test).
4. Timestamps are ISO 8601 in UTC. Fields that take an instant (since, until, snoozedUntil) accept ISO 8601 or epoch milliseconds.
5. Platforms are bluesky, hackernews, github, stackoverflow, devto, reddit, x, youtube, news, linkedin. The field is always called platform (or platforms for a list).
6. Classification: relevance 0 to 100 (relevant is true from 40 up), sentiment is positive, neutral or negative, intents are any of buy_intent, question, complaint, praise, comparison, language is an ISO 639-1 code or null. It is null until the classifier has run. The user can overrule it: PATCH /v1/mentions/{id} with relevant or sentiment (rules/mentions.md), and classification.feedback shows the verdict.
7. A mention's status is the user's triage: open (untouched), ignored (hidden from the feed and every channel), done (handled). Ignored and done mentions are never delivered; only relevant mentions are delivered at all.
8. Errors are { "error": { "code", "message" } }. Branch on code (rules/errors.md). 403 read_only_key means the key cannot write; 402 means the prepaid balance cannot cover the change.
9. Query parameters are camelCase, booleans are true or false, and a list parameter is comma-separated or repeated (platforms=reddit,hackernews). An omitted parameter imposes no constraint.
10. Money: a keyword costs $5 per month, debited daily from a prepaid balance, and every matched mention $0.008, relevant or not. Alerts, digests, reads and analytics are free. Say the cost once when creating the first keyword of a conversation; do not repeat it for every keyword.

What the user asks, what you call

Care

• Confirm with the user before DELETE /v1/keywords/{id} (removes its matches and disables an alert rule that named only it), DELETE /v1/members/{id}, any other DELETE, POST /v1/channels/{id}/rotate-secret (the old secret stops verifying at once) and merging people.
• Do not create API keys unless asked, and hand a new key to the user once without storing it.
• Do not page through the whole feed unprompted: 20 to 50 mentions make a summary. Link the post URL instead of pasting long text.
• When a 402 comes back, the workspace balance is empty or too low: point the user to https://app.mentio.dev/billing rather than retrying.
• When something is not covered here, read https://docs.mentio.dev/llms.txt for the list of doc pages and fetch the one you need as .mdx.

Rule files

Read the one for the area before the first call to it:

• rules/authentication.md: key format, scopes, where keys come from, SDK and CLI use of the same key
• rules/keywords.md: track, pause, widen and remove keywords; matching rules, the per-keyword context and the workspace filters; what a keyword costs
• rules/mentions.md: every search filter, the mention shape, triage and the user's verdicts on the classifier, CSV export
• rules/people.md: the people behind the mentions, tags, notes, mute, merge, saved segments
• rules/alerts.md: rules and channels, daily and weekly digests, webhooks and their signature, testing and the delivery log
• rules/analytics.md: the four reports and the window grammar
• rules/account.md: whoami, the company profile the classifier reads, API keys and their expiry, usage and balance, the team, health
• rules/errors.md: every error code and what to do about it
• rules/cli.md: the mentio CLI, the TypeScript and Python SDKs, the OpenAPI document