https://mcp.arcmira.com/mcp, so there is nothing to install or update. Source: github.com/arcmira/mcp.
Connect
Paste this into your agent. It adds the server and hands you the sign-in:
Every host, step by step: arcmira.com/agent-setup. A host that supports MCP sign-in asks you to approve read access and monitor changes, and, if you are in a team, which account’s credits to use. Any other host sends an account key as
Authorization: Bearer arc_sk_...; prefer signing in, since a key in a config file sits in plaintext. Without either, the server answers 401, and its body carries the sign-up call under error.data.unlock.action, so an agent can create a Free account and reconnect. See Keys and sign-in.
Run your first search
Ask your agent:Search YouTube and podcast transcripts for GPU financing over the last 60 days. Return the relevant quotes with speakers when available, publication dates, transcript sources, and timestamped links. Read the surrounding exchanges before summarizing what the sources imply.The agent calls
arcmira_describe once to load the client reference, then runs a program with arcmira_execute_read. Return structured fields for further analysis and a compact, sourced answer for the reader.
Choose the method for the task
Resolve names before applying entity or channel ID filters. Plain keyword search does not need a resolver call. The MCP client’s argument names differ from the HTTP and SDK names; use the reference returned by
arcmira_describe.
Continue the research
Continue the investment research workflow. You will learn how to follow a search result into competing explanations, inspect the surrounding exchange, verify claims against disclosures, and write a brief with the questions that remain unresolved. Speaker identification and commercial classifications can be incomplete. Check the linked source. Missing a source or a capability your workflow needs? Usearcmira_feedback to describe it, why it matters, and what better support would be worth. Help shape the index.
When you name no time window, the agent uses the last 90 days.
Tools
The server has four tools.
Only
arcmira_describe is read-only, and only arcmira_execute_write is destructive, because it changes monitors you already have. A Premium read only adds. The server declares these annotations; they describe what a tool can do, not what one call costs.
Both execute tools can start a Premium transcript.
arcmira_execute_write also creates and changes monitors and trackers with arcmira.monitors.create, arcmira.monitors.update(id, { paused }), addEntities, addName and attachTrackers; it never deletes. Each feedback call adds a record. Every tool takes an optional intent, the user’s request in at most 300 characters. An empty, failed or truncated result ends with a line naming its call_id for arcmira_feedback.
One program answers one question: it resolves each name to an id, runs the queries, and returns only what the answer needs. Each API call costs what the same HTTP request costs. A program gets 30 seconds, 5 seconds of CPU, 40 API calls and 12,000 characters of output. It runs in an isolate whose only network is a proxy that adds your credential, so the program never holds the key. The read tool reaches GET /v1/*; the write tool adds POST and PATCH under /v1/monitors and /v1/trackers. A plan gate throws the API’s code with an unlock.url the agent relays; see Errors.
Premium transcripts
A Premium transcript is one read inarcmira_execute_read:
pending, the agent reads again later. arcmira.quote(video) shows the price for free.
After a failure, the agent reports why and retries with retry: true only when you ask. If your plan and budget cannot cover the read, the agent says so and links Dashboard → Spending. See Transcripts and Plans and usage.
Not ready for a paid plan? Tell us why and what you would pay for. Feedback is free to give and helps us build what you want.
Monitors from your agent
The agent lists your monitors before it suggests one and asks how you want updates. Thenarcmira_execute_write creates the monitor and follows every entity in one call. It never deletes; it pauses. Delete from the dashboard. A connection made before server 0.9.0 answers insufficient_scope: reconnect Arcmira in the host and allow monitor changes.
The plugin
The plugin adds the server plus skills for sponsor research, company watch, quotes, person research and comparing shows./plugin, Marketplaces, arcmira, Enable auto-update. For other hosts, npx arcmira setup installs the server and skills; see CLI and SDKs.
What we log
Since server 0.8.2, each tool call is recorded in product analytics, linked to your account, so we can fix what fails: the tool, the host, the duration, the first 4,000 characters of the program or the feedback category and note, the agent’s one-lineintent, and the outcome. Credentials and email addresses are redacted first. Results, transcript text and program output are never recorded. Source: src/telemetry.ts.
Older tool names
A host that cached an old tool list getstool_retired, which names the replacement. Server 0.9.0 renamed describe to arcmira_describe and execute to arcmira_execute_read, and prepare_transcript became a Premium read. Server 0.6.0 and earlier had one tool per operation, now one client method each: resolve_entities is arcmira.resolve, search_transcripts is arcmira.search, list_mentions is arcmira.mentions, entity_momentum is arcmira.momentum, count_occurrences is arcmira.occurrences, list_episodes is arcmira.episodes, list_sponsors is arcmira.sponsors, list_recommendations is arcmira.recommendations, index_status is arcmira.status and get_transcript is arcmira.transcript. Since 0.10.0, arcmira.prepare and arcmira.wait return method_retired.