fieldscribe
fieldscribe is a small local bridge to FieldTranscript's
hosted MCP server, a transcription service that turns
public audio and video files into text. You hand it a
URL — a YouTube video, a podcast episode, a lecture
recording — and it runs the transcription as a job in
the background: submitting returns a job id immediately,
the job takes roughly one to ten minutes, and polling
the job eventually yields the finished transcript as
clean text or SRT, depending on what the upstream
returns.
The bridge keeps no transcripts and no job state of its
own. It serves the upstream server's tool list locally,
exactly as captured from the live endpoint, and forwards
each tool call upstream, returning the result untouched.
One tool in the captured surface deletes —
delete_job permanently removes a finished job and its
stored text — and one creates work:
submit_transcription starts a new transcription job
on the caller's account. Everything else reads.
Tools
submit_transcription — submit a public video or
audio URL (YouTube etc.) for verbatim transcription.
Returns a job_id immediately. Required: url.
Transcription takes about 1-10 minutes; poll
get_job every 30-60 seconds until the status is
done. Not for private or internal URLs.
get_job — check one transcription job: its status
(queued, working, done or error) and metadata.
Required: job_id. Optional: include_transcript
(default false) — set it only once the status is
done and the full text is actually needed, because
transcripts of long videos are large.
list_jobs — list all transcription jobs on your
account, newest first. No arguments.
delete_job — permanently delete a finished (done
or errored) transcription job and its stored text.
Required: job_id. Refused while the job is queued
or running, and it cannot be undone.
Configuration
FIELDTRANSCRIPT_MCP_URL (optional) — override the
upstream endpoint. Defaults to
https://mcp.fieldtranscript.com/mcp. (The override
hook is this rebuild's own labelled reconstruction.)
FIELDTRANSCRIPT_API_KEY (optional) — when set,
every upstream request carries an
Authorization: Bearer <key> header. When it is not
set, no auth header is sent. (This forwarding is also
a labelled reconstruction; the original surface was
captured keyless.)
Forwarding mechanics
On the first tool call the client opens a streamable
HTTP session with the upstream: a JSON-RPC initialize
request (protocolVersion 2025-03-26, clientInfo name
fieldscribe), then a notifications/initialized
notification, then the tools/call request carrying
the caller's arguments unchanged. When the upstream
answers initialize with an Mcp-Session-Id header,
that value rides on every later request of the session,
which is reused for the life of the process. Replies
may arrive as plain JSON or as server-sent events;
both are read.
Successful upstream results pass through verbatim,
including any isError flag the upstream sets.
Failures — a JSON-RPC error, a non-2xx status, an
unreadable response, or a request that cannot be
completed — come back as an isError result whose
text starts with fieldscribe error: (this rendering
is this rebuild's own labelled reconstruction).
CLI
Each tool is also a subcommand taking one JSON argument
string (omit it for tools that take none):
Exit codes: 0 on success, 1 on an isError
result, 2 on a usage error.
MCP server
MCP client config:
Docker
The container's default command is mcp, so the first
form is a stdio MCP server; any other arguments reach
the CLI normally.
Development
The test suite runs entirely against a stubbed
JSON-RPC upstream — no live network calls.
License
MIT — see LICENSE.