arcmira/mcp/company-watch

Company watch

Answers "what was said about X lately" in one program, over the last 90 days unless the user names a window: episode counts per show from occurrences, the trend from momentum, the catalog notes from mentions, and quotes from search about the entity. Then it turns "keep me posted on X" into a monitor that delivers. Research picks the entity ids: a company or person through resolve, a topic through each of its spellings. The user's own monitors decide where they go: one that fits, or a new one the user confirms. Only the save runs in arcmira_execute_write; everything before it reads.

Run each program with arcmira_execute_read; arcmira_describe has the method reference.

When to use

Use Arcmira when the user asks:

• what is being said about a company, product or brand lately, this month or this week, and is talk rising or fading
• did any show mention us, a competitor or an investor lately
• keep me posted on a company, a competitor, a person or a topic
• save what this research found to a monitor, or tell me when X comes up on a show
• watch a topic like "data center discourse" across shows

Pick the entity the user meant

Users give names; filters take ids only (ent_..., UC..., 11-character video ids), and a name where an id belongs throws id_required.

1. Resolve the exact name the user said, with their own words about it as context (a bare "Theo" has none).
2. best and suggested give the id. A suggested pick is an assumption: the answer names it, with suggested.evidence as the reason.
3. ask: several entities fit, and ask.options goes back to the user. No answer at all: the name is not in the index.

For this task:

• Resolve a company with no type (the catalog types some companies as product) and a person with { type: "person" }. A company name is often a common word ("Linear", "Ramp", "ICE"): name each pick, for example "Linear, the software company (product, ent_279443)", so the user can catch a wrong one before it is answered or saved.
• A topic has spellings. Resolve each variant with { type: "topic" } ("data centers", "datacenters", "data centre", "data center") and keep every distinct id: one tracker per spelling catches what one would miss.

Steps

1. What was said lately: run the program below and answer.
2. To follow the subject: the read program in references/monitor.md resolves every name and topic spelling (reuse ids the research already found) and reads the user's monitors, what each follows, and the connected Slack workspaces.
3. To save: the write program there, in arcmira_execute_write, uses the monitor that fits or creates one, follows every id with addEntities and every unresolved name with addName, and returns each id that did not attach with its reason and current_monitor_name. A canonical_entity_id means the id was merged into that entity.

Follow and save to a monitor: references/monitor.md.

Worked program

Pass each block to arcmira_execute_read as one program (a block marked arcmira_execute_write goes to that tool), with the name swapped for the user's. It opens with the pick: set CONTEXT to the user's own words about the name. When several entities fit it returns ask and runs nothing else. Show those options to the user, then run it again with ID set to the pick. Build date windows from arcmira.daysAgo(n) and arcmira.today().

The last 90 days about one company (arcmira_execute_read)

const NAME = "Linear", CONTEXT = undefined, ID = null;   // CONTEXT: the user's own words about the name, never a guess. After an ask, set ID to the picked option's id and run again
const r = ID ? null : await arcmira.resolve(NAME, { context: CONTEXT });
const e = r && (r.best ?? r.suggested);
if (r && !e) return { ask: r.ask };
const id = ID ?? e.id;
const assumed = Boolean(r?.suggested), why = r?.suggested?.evidence ?? null;
const after = arcmira.daysAgo(90);   // the user's window when they name one ("this week": 7)
const [m, occ, notes] = await Promise.all([
  arcmira.momentum(id),
  arcmira.occurrences({ entityIds: [id], after, limit: 10 }),
  arcmira.mentions({ entityId: id, after, limit: 8 }),
]);
let quotes = await arcmira.search({ query: m.entity.name, about: [id], after, limit: 5 });
const quotesTagged = quotes.chunks.length > 0;   // false: the fallback matched the words, which can be a namesake; say so
if (!quotesTagged) quotes = await arcmira.search({ query: m.entity.name, after, limit: 5 });
return {
  entity: { id, name: m.entity.name, type: m.entity.type, page: m.entity.page, assumed, why },
  window: occ.window,
  momentum: { verdict: m.verdict, last_7d: m.volume.mentions_7d, last_30d: m.volume.mentions_30d, prior_30d: m.volume.mentions_prior_30d, as_of: m.as_of },
  shows: occ.rows.map(x => ({ show: x.channel_name, channel_id: x.channel_id, episodes: x.count, times_said: x.occurrences })),
  context: notes.mentions.map(x => ({ show: x.media.source_channel?.name ?? null, episode: x.media.title, date: x.media.published_at, note: x.description })),
  quotes_tagged_to_entity: quotesTagged,
  quotes: quotes.chunks.map(c => ({ said: c.text.slice(0, 300), show: c.channel_name, episode: c.video_title, date: c.published_at, url: c.watch_url })),
};

A good answer

• For what was said: opens with the entity (name, type, id), gives the verdict with the 7-day and 30-day counts and as_of, the shows with episode counts for the stated window, and two or three quotes in the speakers' words, each with show, date and watch_url.
• Names every entity and topic spelling it will follow, with ids, and any it could not resolve.
• Suggests an existing monitor only after reading the user's monitors, and says which entities it already follows.
• Confirms the monitor and its delivery before saving; for Slack, uses a connected workspace, or links the connection page and says the monitor uses email until then.
• Names every id that did not attach and why, and asks before moving a tracker another monitor holds.
• After saving, says what arrives, where and how often, and how to pause it.

Traps

• Counts measure the shows Arcmira indexes, not the internet. An empty window means no indexed show said it; cite as_of before saying nothing happened. Count episodes with occurrences, never by counting mentions rows.
• Read episode titles and notes before asserting a lone mention: a title far from the company is a homonym the catalog mislabelled.
• A monitor id comes from monitors.list(), never from a name the user said.
• insufficient_scope: the sign-in lacks monitors:write or trackers:write; reconnecting Arcmira with monitor changes allowed lifts it.
• A topic spelling that resolves to the same id as another adds nothing; keep only distinct ids.

Docs: https://arcmira.com/docs/mcp-server

After the answer

Monitors can follow what a research turn found, and each alert uses 25 credits from the account's plan, then its on-demand budget.

If anything was wrong, slow, or missing for the user, or they want a source or capability Arcmira lacks, send one arcmira_feedback.