Arcmira: YouTube Transcript Search

What Arcmira: YouTube Transcript Search does

Arcmira: YouTube Transcript Search gives agents access to Arcmira’s indexed YouTube and podcast transcript data through MCP. It can locate passages containing a phrase, return timestamped transcript lines, identify mentions of people, companies, products, or topics, and examine where an entity appears across channels.

The service also supports media-appearance counts, mention-momentum comparisons, identification of organic endorsements and ad reads, recurring channel sponsors, and monitoring for new YouTube mentions. Search results can include passage text, available attribution, publication dates, source links, timestamps, and related classification fields.

How it works

Connect an MCP client to https://mcp.arcmira.com/mcp. The client first calls arcmira_describe to load the relevant method reference, then sends code to arcmira_execute_read or arcmira_execute_write. The code runs with arcmira and ArcmiraError available and returns both printed output and the program’s return value.

Use the client methods according to the research question. arcmira.search finds transcript passages, arcmira.transcript reads surrounding lines, arcmira.mentions finds entity references, and arcmira.occurrences counts appearances. arcmira.momentum, arcmira.recommendations, and arcmira.sponsors cover trend comparisons, commercial classifications, and sponsor history. Resolve names before applying entity or channel ID filters; plain keyword searches do not require resolution.

Setup and configuration

Add the remote MCP URL to a compatible host and authenticate either through the host’s OAuth flow or with a bearer key. OAuth-capable hosts can sign in and approve permissions in Arcmira. Clients without that flow can send an account key in the Authorization: Bearer header. The key can be obtained from Arcmira, including through its email-based signup API.

For example, Claude Desktop or another HTTP-capable configuration can use the remote URL with an Authorization header containing ARCMIRA_API_KEY. Claude Desktop, Codex, and Cursor setup examples are available through the service’s agent setup instructions.

Tools and capabilities

• arcmira_describe returns the client reference, method guidance, arguments, return fields, examples, usage rules, error codes, and documentation links. It does not bill.
• arcmira_execute_read runs read-oriented programs, including transcript searches, transcript retrieval, and monitor reads.
• arcmira_execute_write additionally creates and updates monitors, adds tracked entities or names, and attaches trackers. It can overwrite monitor settings or pause monitors; it does not delete data.
• arcmira_feedback records experience feedback such as incorrect entities, bad data, missing results, slow responses, or confusing behavior.

Execution is limited to 30 seconds, 40 API calls, 12,000 total output characters, 3,000 characters per string, and 100 items per array. Searches can return up to 20 passages, so returning only the fields needed for later analysis helps avoid truncation.

Limitations and notes

Reads are metered. Search passages beyond the first five, caption windows, and Premium transcription use credits, followed by the configured on-demand budget. Premium transcription can be initiated through either execution tool and reports its charge. Account plan, scopes, and budget affect available results.

Speaker identification, source coverage, and organic-versus-sponsored classifications can be incomplete or incorrect. Verify important findings against the linked video or podcast source. The public plugin files use Apache-2.0, while the hosted research service is proprietary.