| name | mediamtx-docs |
| description | Search and fetch official MediaMTX documentation from mediamtx.org: kickoff (introduction, install, upgrade), features (architecture, publish, read, record, playback, authentication, hooks, metrics, forward, proxy, performance, rtsp-specific-features, rtmp-specific-features, webrtc-specific-features, srt-specific-features, absolute-timestamps, always-available, on-demand-publishing, decrease-packet-loss, extract-snapshots, remuxing-reencoding-compression, start-on-boot, embed-streams-in-a-website, expose-the-server-in-a-subfolder, logging, configuration), publish + read howtos (ffmpeg, gstreamer, obs-studio, python-opencv, golang, unity, raspberry-pi-cameras, rtsp, rtmp, hls, srt, webrtc, rtp, mpeg-ts), references (configuration-file, control-api). Use when the user asks to look up a MediaMTX config option, find a control-api endpoint, verify supported protocols, check auth modes (internal/HTTP/JWT), or read mediamtx howtos.
|
| argument-hint | [query] |
MediaMTX Docs
Context: $ARGUMENTS
Quick start
- Find a config key / API endpoint / feature: -> Step 2 (
search --query <term>)
- Read a how-to in full: -> Step 4 (
fetch --page <publish/X or read/X>)
- Read one section: -> Step 3 (
section --page <page> --id <anchor>)
- Prime cache for offline: -> Step 5 (
index)
When to use
- User asks "how do I push RTSP to MediaMTX?" or "how do I configure JWT auth?" or "what control-api endpoints exist?"
- Need to verify a MediaMTX config key or CLI flag before recommending it.
- Need the canonical mediamtx.org URL to cite.
- Before wiring up a publisher/reader, confirm the protocol endpoint and default port (RTSP 8554, RTMP 1935, HLS 8888, WebRTC 8889, SRT 8890, API 9997, Metrics 9998, pprof 9999).
- For the running-server wrapper + API calls, see the
mediamtx-server skill.
Step 1 — Know the page catalog
uv run ${CLAUDE_SKILL_DIR}/scripts/mtxdocs.py list-pages
Top-level families:
| Family | When to pick |
|---|
kickoff/* | Install, upgrade, first-run. |
features/* | Core concepts (publish, read, record, playback, auth, hooks, metrics). |
features/<proto>-specific-features | Per-protocol options (rtsp-specific-features, rtmp-specific-features, webrtc-specific-features, srt-specific-features). |
features/configuration + features/control-api + features/logging + features/metrics | Operational — the knobs you actually turn in production. |
references/configuration-file | Authoritative list of every YAML option. Start here for config questions. |
references/control-api | Authoritative list of every /v3/* endpoint. |
publish/<client> | How to SEND a stream in via ffmpeg / gstreamer / OBS / Python / Go / Unity / etc. |
read/<client> | How to PLAY a stream out via ffmpeg / gstreamer / VLC / OBS / browsers / etc. |
github-readme, github-mediamtx.yml, github-apidocs | Raw upstream files — the ground truth when mediamtx.org is behind. |
Read references/pages.md for the full catalog.
Step 2 — Search first
uv run ${CLAUDE_SKILL_DIR}/scripts/mtxdocs.py search --query "recordFormat" --limit 5
uv run ${CLAUDE_SKILL_DIR}/scripts/mtxdocs.py search --query "jwt" --page features/authentication
uv run ${CLAUDE_SKILL_DIR}/scripts/mtxdocs.py search --query "/v3/paths" --page references/control-api
Use --regex for anchored patterns. Each hit prints page:line, nearest heading, canonical URL, and ±3 lines of context.
Step 3 — Read one section
When a hit surfaces [§anchor]:
uv run ${CLAUDE_SKILL_DIR}/scripts/mtxdocs.py section --page features/authentication --id internal-users
uv run ${CLAUDE_SKILL_DIR}/scripts/mtxdocs.py section --page references/configuration-file --id paths
--id also accepts a heading keyword as fallback.
Step 4 — Fetch a whole page
Use sparingly — howto pages are compact enough to fetch whole when you actually need the full recipe:
uv run ${CLAUDE_SKILL_DIR}/scripts/mtxdocs.py fetch --page publish/ffmpeg
uv run ${CLAUDE_SKILL_DIR}/scripts/mtxdocs.py fetch --page read/web-browsers
uv run ${CLAUDE_SKILL_DIR}/scripts/mtxdocs.py fetch --page github-mediamtx.yml
--format json for structured handoff.
Step 5 — Prime cache
uv run ${CLAUDE_SKILL_DIR}/scripts/mtxdocs.py index
Fetches every known page (~70 URLs) at 0.3s each. Override cache dir via MEDIAMTX_DOCS_CACHE. Clear with clear-cache.
Gotchas
- MediaMTX does NOT transcode — it remuxes only. If a publisher pushes HEVC into a path and a reader asks for WebRTC in the browser, the reader will FAIL (browsers don't universally support HEVC in WebRTC). The docs page
features/remuxing-reencoding-compression is explicit: for transcoding, chain an external ffmpeg publisher. When in doubt, publish as H.264 + Opus.
- The binary is now
mediamtx (was rtsp-simple-server). Repo moved from aler9/rtsp-simple-server -> bluenviron/mediamtx. Old Docker tags and install scripts still float around the internet; ignore them.
- Default ports you'll see quoted in docs: RTSP 8554, RTSPS 8322, RTMP 1935, RTMPS 1936, HLS 8888, WebRTC 8889, SRT 8890, Control API 9997, Metrics 9998, pprof 9999, Playback 9996. Each is a separate listener; disable any you don't need in
mediamtx.yml.
- Three auth backends, exclusive.
internal (users in YAML under authInternalUsers), http (POST to an external endpoint), jwt (verify a bearer token via JWK URL or static secret). You can't mix them globally — but per-path permissions still apply. See features/authentication.
- Auth actions:
publish, read, playback, api, metrics, pprof. Default allowed users have all which is shorthand for the union. Lock down production before exposing port 9997.
- Hooks fire on path events, not on connection events.
runOnInit runs when the path is first registered, not when a client connects. runOnDemand is what you want for lazy pipelines. Read features/hooks.
always-available vs on-demand-publishing. Set sourceOnDemand: true to start source: only when readers appear (saves bandwidth). sourceOnDemandStartTimeout + sourceOnDemandCloseAfter control how long MediaMTX waits / keeps source alive after the last reader.- Control API lives at
/v3/*, NOT /v2/* or . Older tutorials reference v2; the v3 namespace is the current stable. Full endpoint list is on .
Examples
Example 1 — "What do I put in mediamtx.yml to enable JWT auth?"
uv run ${CLAUDE_SKILL_DIR}/scripts/mtxdocs.py search --query "jwt" --page features/authentication --limit 5
uv run ${CLAUDE_SKILL_DIR}/scripts/mtxdocs.py section --page features/authentication --id jwt
Example 2 — "How do I publish with ffmpeg?"
uv run ${CLAUDE_SKILL_DIR}/scripts/mtxdocs.py fetch --page publish/ffmpeg
Example 3 — "What's the full list of control-api endpoints?"
uv run ${CLAUDE_SKILL_DIR}/scripts/mtxdocs.py fetch --page references/control-api
Example 4 — "What does runOnDemandCloseAfter default to?"
uv run ${CLAUDE_SKILL_DIR}/scripts/mtxdocs.py search --query "runOnDemandCloseAfter" --page github-mediamtx.yml --limit 3
(Upstream YAML has inline defaults — faster than the HTML docs here.)
Example 5 — "Record to disk, rotate every 15 min, keep 4 hours"
uv run ${CLAUDE_SKILL_DIR}/scripts/mtxdocs.py search --query "recordSegmentDuration" --page features/record
uv run ${CLAUDE_SKILL_DIR}/scripts/mtxdocs.py search --query "recordDeleteAfter" --page features/record
Example 6 — "WebRTC publish from a browser — what's the WHIP URL?"
uv run ${CLAUDE_SKILL_DIR}/scripts/mtxdocs.py fetch --page publish/web-browsers
Troubleshooting
Error: unknown page: foo
Cause: The name isn't in the catalog.
Solution: Run list-pages. Common mistakes: using record instead of features/record; api instead of references/control-api.
Error: urlopen error [SSL: CERTIFICATE_VERIFY_FAILED]
Cause: System certificate store is out of date (usually macOS Python).
Solution: Run /Applications/Python\ 3.x/Install\ Certificates.command or set SSL_CERT_FILE. Don't patch SSL verification off.
Search returns zero hits for a config key
Cause: Key might be new in main but not yet in the published docs.
Solution: Search against the upstream YAML — search --query "<key>" --page github-mediamtx.yml. The upstream file is the source of truth.
Results look like broken-up tables
Cause: The HTML extractor flattens cell borders into pipes but can't reconstruct multi-line cells.
Solution: The search-hit header prints the canonical URL. Open it directly in a browser for the authoritative view.
Cache stale after MediaMTX release
Solution: clear-cache then index. The GitHub fallback pages already track main — if you want a specific release tag, edit the github-* URLs in the script's PAGES dict (or fetch directly with urllib).
Reference docs
