Arcmira is a search engine for the spoken web: an index of YouTube and podcast transcripts covering mentions, momentum, sponsors, and on-air recommendations.
Agen.co connects to Arcmira through Arcmira's own hosted MCP server as an Official connector. Each user authorizes with their own Arcmira account. Agents research the transcript index at that account's plan level and, when the user allows it, can also create and change the user's Arcmira monitors.
Prerequisites
Prerequisites
- An Arcmira account. No app needs to be registered with Arcmira — the MCP server registers Agen.co as a client automatically, so there is no Client ID or Client Secret to copy. A reader with no account can create one from the consent screen's sign-up flow.
This server runs programs, not one tool per operation
This server runs programs, not one tool per operation
Arcmira's MCP server uses a code-execution pattern. arcmira_describe returns the Arcmira client's reference — methods, arguments, and worked examples — and arcmira_execute_read and arcmira_execute_write run a short JavaScript program written against that client inside a sandbox, returning what the program prints or returns. An agent calls arcmira_describe once to learn the client, then writes and runs programs against it. Older descriptions of Arcmira's server list a describe and execute pair, or separate tools such as searching transcripts. Those names are retired, and a call to one returns an error naming its replacement.
In the Agen.co portal, go to Connectors → My connectors and click Add connector.
In the Select connector drawer, search for
Arcmiraand select it. The Add Arcmira panel opens.Fill in the field:
Field Required Description Instance Slug Yes Prefixes each imported tool as slug__tool, so several instances of the connector can coexist. Prefilled witharcmira, or with the next free slug such asarcmira-2if Arcmira is already connected. Use lowercase kebab-case. You can change it later from the connector's settings.Click Connect. Arcmira opens in a new tab, and the panel shows Waiting for authorization — complete it in the opened tab…
On the Connect MCP OAuth Client screen, which reads MCP OAuth Client is asking for access to your Arcmira account, sign in if prompted. The screen asks you to approve read access and monitor changes. If you belong to an Arcmira team, pick the account whose plan and credits the connection uses under Uses the credits of. Click Allow.
Back in Agen.co, the panel switches to Select the tools to import from Arcmira. All four tools are toggled on; turn off any you don't want to import, then click Add.
The connector is created and its tools imported only when you click Add. If you close the panel before that, nothing is saved, even though the Arcmira tab reported success.
Adding the connector doesn't authorize anyone's tool calls yet. The first time each user's agent calls an Arcmira tool, the call returns an authorization link instead of a result. The user opens the link and approves the same way, including whether to allow monitor changes. After that, the user's tool calls run against their own Arcmira account, at that account's plan level and using its credits.
Once connected, Arcmira appears under My connectors with 4 tools:
| Tool | What it does |
|---|---|
| arcmira_describe | Returns the Arcmira client reference: methods, arguments and return fields, worked example programs, error codes, and the rules for credits and monitors. Can be narrowed to one topic, such as sponsors or monitors. Reads nothing from the index and uses no credits. |
| arcmira_execute_read | Runs a short JavaScript program against the Arcmira client and returns what it printed or returned. This is how research runs: searching transcripts, mentions, momentum, sponsors, recommendations, coverage, Premium transcripts, and listing the user's monitors. It doesn't change monitors, but a Premium transcript read can use credits. |
| arcmira_execute_write | Runs programs like arcmira_execute_read, plus monitor changes: creating a monitor, following entities or names in it, changing how and how often it sends updates, and pausing it. It never deletes; a monitor is paused instead. |
| arcmira_feedback | Sends Arcmira a feedback record about a wrong, slow, missing, or confusing result. Uses no credits. |
Grant arcmira_execute_write only to agents that should change users' monitors. The other three tools never change monitors or other account settings.
Enabling the Arcmira connector isn't enough on its own. Tool calls remain denied until you create a policy that grants access to the specific tools you want to expose.
- Monitor changes need the user's permission.
arcmira_execute_writeworks only for users who allowed monitor changes when they authorized. For a user who didn't, monitor changes fail with a missing-permission error until they authorize again and allow it. Reads don't need this permission. - Premium transcripts use credits. Premium, speaker-labeled transcripts need a paid Arcmira plan. When a video isn't transcribed yet, a Premium read transcribes it, using credits from the account's plan first and then its on-demand budget, without asking the user to confirm — the on-demand budget set in the Arcmira dashboard is the spending limit. Arcmira charges for the whole video, never charges twice for a video the account already has, and lets the agent check the cost first at no charge.
- Other usage is metered too. Rows past the first few in a request, sponsor rows, and each monitor alert also use credits; creating a monitor is free. See Arcmira pricing for current rates.
- Plan gates are separate from permissions. Data the account's plan doesn't include — such as Premium transcripts, sponsor rows, or the newest media on the Free plan — comes back as an error with a link to unlock it, not the data. Allowing monitor changes doesn't unlock plan features. When the plan's credits and the on-demand budget run out, reads that need credits are refused until an account admin raises the budget or upgrades the plan.
- Each program is bounded. A program can run for up to 30 seconds and make up to 40 Arcmira API calls, and its output is capped at 12,000 characters. A result over the cap comes back truncated, and the agent has to ask for fewer fields or a smaller page.
If your Arcmira connector still lists the retired describe and execute tools, their calls return an error naming the replacement. Reconnect the connector to refresh the tool list, then update any policy that granted describe or execute so it grants the new tools instead — a policy that names the old tools doesn't cover the new ones.