bi-mcp
This server provides read-only inspection of a Blue Iris NVR — cameras, alerts, clips, PTZ, logs, and config — mostly answering plain-English questions rather than changing anything.
Cameras & groups — list all cameras with online state, stream bitrate/FPS/resolution, trigger/alert counts, and last alert time.
Camera configuration — per-camera config (motion sensitivity, AI zones, recording mode, stream paths, schedule/profile flags) and live motion/post-trigger settings; deep views require admin.
Alerts — list recent alerts with AI memo (object, confidence, license plate), zones, memos, and clip path; filter by date range, view, and memo substring.
Clips — list recorded clips (continuous + alert-linked) with duration, resolution, flags, memo; supports day-tiles and date filters.
Images — fetch a live JPEG snapshot of any camera, or the stored alert image (optionally with AI detection overlay) resolved by time.
Forensics — get detailed clip info (resolution, duration, active profile/schedule/zones), explain an alert's action chain with a ±2-minute log cross-reference, and fetch AI object tracks (currently broken on BI 5.9.9.71).
Registry / .reg parsing — extract trigger-zone polygons, per-class AI confidence thresholds, per-preset alert-skip flags, ONVIF event handlers, and alert action definitions not exposed by the BI JSON API.
Action sets — decode OnTrigger/OnReset action rows (types, commands, PTZ presets, protocol, bitmasks) into readable form.
Action audit — surface cross-camera action-row outliers for review (informational, not bug detection).
PTZ — read current position, preset list, preset_map, and active preset for PTZ cameras.
System — status snapshot (active profile, schedule, CPU/RAM/disk, uptime, DIO, warnings), session info (BI version, user capabilities, available profiles/schedules/streams), and system config (admin).
Logs & timeline — query BI system log with filters (time, camera/obj, levels, substring/regex), and get a 24-hour activity timeline per camera.
All tools are read-only; a few admin-gated ones degrade or error cleanly without admin credentials. Mutating tools exist but aren't part of this schema.
Click on "Deploy Server".
Wait a few minutes for the server to deploy. Once ready, it will show a "Started" state.
In the chat, type
@followed by the MCP server name and your instructions, e.g., "@bi-mcp What cameras are offline right now?"
That's it! The server will respond to your query, and you can continue using it as needed.
Here is a step-by-step guide with screenshots.
bi-mcp: Blue Iris MCP Server
Ask Claude about your security cameras in plain English. It looks at your live Blue Iris install and answers, instead of guessing. What cameras are online, why an alert fired, what the AI saw, how a preset is configured.
What is this?
Blue Iris is a Windows NVR app for IP security cameras. It records footage, runs motion and AI detection, and manages PTZ cameras.
The Model Context Protocol is a standard way for AI assistants like Claude to call tools on your own computer.
bi-mcp is an MCP server that connects the two. Claude gets a set of read-only tools (and optional control tools) for inspecting your Blue Iris install. You talk to Claude, Claude talks to Blue Iris.
Some example questions you can ask once it's set up:
What cameras are offline right now?
Why did the front door camera fire an alert at 3:14 PM?
Show me the trigger zones on SecCam_3.
What's the current PTZ preset on the pan-tilt camera?
18 read-only tools register by default. 6 control tools (rename a
camera, switch profile, recall a PTZ preset, export a clip, and a couple
more) register only when you explicitly opt in by setting
BI_MCP_ALLOW_MUTATIONS=1. The project is MIT-licensed.
Compatibility note: bi-mcp is built and tested against Blue Iris 5.x (specifically 5.9.9.71). Blue Iris 6 is out but has not been tested. Some tools call undocumented BI endpoints whose response shapes may have changed in 6.x. If you run bi-mcp against Blue Iris 6 and something works (or breaks), open an issue so I can update this note.
Related MCP server: Atlassian Jira MCP Server
Prerequisites
You need three things before installing.
First, Blue Iris 5.x running on Windows with the web server enabled (see the compatibility note above about Blue Iris 6). In Blue Iris that's under Settings, then Web server, then check "Enabled".
Second, Python 3.10 or newer on the machine that will run bi-mcp. To
check what you have, open a terminal and run python --version. If it's
missing or too old, install from
python.org.
Third, a Claude app. Either Claude Desktop, which most users want, or Claude Code, which is the terminal CLI.
bi-mcp itself can run on the Blue Iris box or on any other computer on the same LAN.
Install
bi-mcp is a command-line tool. The recommended way to install command-line
Python tools is with pipx, which gives each tool its own isolated
environment so they don't interfere with each other.
Open a terminal. On Windows that's PowerShell from the Start menu, on macOS it's Terminal from Spotlight, on Linux it's whatever terminal you already use.
Then run one of these:
# Most common, works on macOS, Linux, and Windows
pipx install bi-mcp
# Faster alternative, if you have uv (https://docs.astral.sh/uv/)
uv tool install bi-mcp
# Try it once without installing anything permanent
uvx bi-mcp-server checkYou can also install straight from the GitHub source if you want the latest unreleased changes:
pipx install git+https://github.com/whoamiTM/bi-mcp
# or with uv:
uv tool install git+https://github.com/whoamiTM/bi-mcp
# or one-shot:
uvx --from git+https://github.com/whoamiTM/bi-mcp bi-mcp-server checkIf you don't have pipx, install it with python -m pip install --user pipx followed by python -m pipx ensurepath. Close and reopen your
terminal so the new command lands on your PATH.
After installing, confirm the command works:
bi-mcp-server --helpIf you want to edit the source instead of installing a released version, see For contributors further down.
Create a Blue Iris user for bi-mcp
Don't point bi-mcp at your admin account. Create a dedicated low-privilege
user that bi-mcp will log in as. In Blue Iris, go to Settings then Users
and click the + button to add a user. Name it mcp-readonly (or
whatever you prefer) and set a password you'll paste into the config in
the next step. Set Access to "Local + LAN", leave Admin unchecked, and
check the PTZ and Clips boxes. Under Camera groups, tick the
groups you want Claude to be able to see.
The Security section further down has the full permission table and explains the optional admin user that a handful of tools need.
Quickstart with Claude Code
The simplest way to register an MCP server with Claude Code is the
claude mcp add command. From any terminal:
claude mcp add --transport stdio -s user \
-e BI_HOST=192.168.1.10 \
-e BI_PORT=81 \
-e BI_USER=mcp-readonly \
-e BI_PASS=your-password-here \
bi-mcp \
-- bi-mcp-serverReplace 192.168.1.10 with your Blue Iris box's LAN IP, and the user
and password with the ones you just created. The -s user flag makes
bi-mcp available in every project, not just the current directory.
Then restart Claude Code. Run /exit in any active session, then launch
claude again. In a fresh session, ask:
List my Blue Iris cameras.
Claude should call bi_list_cameras and show your camera list.
Editing the config file by hand
If you'd rather edit the config directly, it lives at ~/.claude.json on
every platform. Add a bi-mcp entry under mcpServers:
{
"mcpServers": {
"bi-mcp": {
"command": "bi-mcp-server",
"env": {
"BI_HOST": "192.168.1.10",
"BI_PORT": "81",
"BI_USER": "mcp-readonly",
"BI_PASS": "your-password-here"
}
}
}
}Same restart procedure as above.
Quickstart with Claude Desktop
I built bi-mcp against Claude Code and haven't tested it on Claude Desktop. The MCP protocol is the same so it should work the same, but the instructions below are lighter than the Claude Code path. PRs welcome if anything's off.
The friendliest way to edit the config is from inside Claude Desktop itself. Open the Claude menu, go to Settings, click the Developer tab, then click "Edit Config". Claude Desktop creates the file if it doesn't exist yet and opens it in your default editor.
If you'd rather find the file on disk, it lives at:
macOS:
~/Library/Application Support/Claude/claude_desktop_config.jsonWindows:
%APPDATA%\Claude\claude_desktop_config.json
Add a bi-mcp entry under mcpServers using the same JSON shape as the
Claude Code block above.
After saving, fully quit Claude Desktop and reopen it. Closing the window is not the same as quitting. Use the app menu's Quit option.
If you want a walkthrough that uses Claude Desktop with a different MCP server (the official filesystem one) as a template, Anthropic's user quickstart at modelcontextprotocol.io/quickstart/user covers the same Settings → Developer → Edit Config flow in more detail.
When the first run doesn't work
The most common cause is invalid JSON in the config file. Trailing commas, mismatched quotes, that kind of thing. Paste the file through a JSON validator if you're not sure.
The next most common cause is not actually restarting the Claude app.
For Claude Code, run /exit and then claude again. For Claude Desktop,
use the app menu's Quit option. Closing the window isn't enough.
If both of those check out, run the CLI smoke test from a terminal:
BI_HOST=192.168.1.10 BI_USER=mcp-readonly BI_PASS=... bi-mcp-server checkA passing run prints OK — connected to Blue Iris 5.x.y.z at HOST:PORT, N cameras found. If you see that, the server itself works and the
problem is on the Claude side: config not loaded, wrong file, app not
fully restarted. If the smoke test fails, jump to
Troubleshooting further down.
You can also call individual tools from the CLI to confirm they return real data:
bi-mcp-server bi_list_cameras
bi-mcp-server bi_get_camera_config --short=SecCam_3And for poking at tool schemas interactively, the MCP Inspector opens a browser UI:
npx @modelcontextprotocol/inspector uvx --from . bi-mcp-serverConfiguration reference
bi-mcp reads its settings from environment variables. The env block in
the Claude config above sets them, so you don't need a separate file.
Variable | Required | Purpose |
| yes | LAN IP or hostname of your Blue Iris box |
| yes | Blue Iris web-server port (default |
| yes | The low-privilege user you created |
| yes | That user's password |
| no | Optional admin user for admin-gated tools (see Security) |
| no | Admin user's password |
| no | Set to |
| no | Set to |
You can also put these in a .env file in the directory you launch the
server from, which is handy for the CLI smoke tests. The .env.example
file in this repo shows the format.
For contributors (editing the source)
git clone https://github.com/whoamiTM/bi-mcp
cd bi-mcp
uv sync
uv run bi-mcp-server checkIf you'd rather use plain pip, note that the development dependencies are
declared as a PEP 735 dependency group rather than as an extra, so they need
their own install step and a pip new enough to understand --group. Running
pip install -e .[dev] instead will quietly install nothing and exit 0,
leaving you without pytest.
python -m venv .venv
source .venv/bin/activate
python -m pip install --upgrade "pip>=25.1"
pip install -e .
pip install --group dev
pytest tests/ -qTool reference
18 read tools register by default. The 6 mutating tools register only when
BI_MCP_ALLOW_MUTATIONS=1. Pass raw=true on any tool to get the unshaped
Blue Iris JSON (on bi_list_alerts/bi_list_clips/bi_list_log an explicit
limit still caps the list).
For the canonical reference with routing tables, error taxonomy, and mutation-safety rules, see AGENTS.md.
Tool | Backing BI cmd | Admin? | Mutating? | Purpose |
|
| Active profile, schedule, CPU/RAM/disk, uptime, DIO, warnings. | ||
|
| BI version/capabilities, available profiles/schedules/streams. | ||
|
| ✓ | Archive/schedule/manrecsec + any DIO/MQTT inline. | |
|
| All cameras and groups: online state, trigger counts, stream health. | ||
|
| (deep) | Per-camera config (deep w/ admin, shallow w/o). | |
|
| ✓ | Live | |
|
| Current JPEG frame, returned as an MCP image block (renders inline in image-aware clients) plus base64. Useful for live coverage cross-reference and PTZ preset framing checks. | ||
|
| Stored alert image (the saved frame, not a live one), resolved by camera + optional | ||
|
| Recent alerts with AI memo, classification, zones, clip path. | ||
|
| Forensic detail for one clip. | ||
|
| Recorded clip inventory; complementary to | ||
|
| Activity timeline (alert/clip spans) for a camera over a window; defaults to the last 24h. | ||
|
| Current PTZ position, preset list, lock state. | ||
|
| ✓ | Recent BI system log entries. | |
| (file parser) | Parse | ||
| (file parser) | Decoded semantic view of | ||
| (file parser) | Cross-camera cohort-divergence report. Surfaces action-row outliers (values that differ from the cohort majority; fields with no strict majority appear as entries with | ||
|
| ✓ | Diagnose one alert: alert facts, per-row filter decode, comparator verdicts for compound/threshold/wait/cross-zone cases, and a ±2-minute log cross-reference of what BI actually did. Refuses alerts older than 24h by default (override with | |
|
| ✓ | ✓ | Fire a synthetic motion trigger (mutations flag). |
|
| ✓ | Recall a PTZ preset 1-20 (mutations flag). | |
|
| ✓ | ✓ | Switch active profile (mutations flag). |
|
| ✓ | ✓ | Async MP4/AVI/WMV export from a clip range (modes: create / status / list). Requires BI user |
|
| ✓ | Set memo (≤35 chars) and/or flag bits on one alert or clip @record. Auto-preserves | |
|
| ✓ | ✓ | 10 ops: rename, hide, enable, audio, output, manrec, pause, profile+lock, reset, reboot. All verify post-write. |
Enabling mutating tools
Set BI_MCP_ALLOW_MUTATIONS=1 in .env (or in the MCP config env block).
With the flag off, the mutating tools are not registered at all, so the
MCP tool list stays clean. Read AGENTS.md § Mutation patterns before
flipping the flag.
bi_get_reg and the .reg parser
bi_get_reg parses Blue Iris's binary .reg camera exports to surface
settings the JSON API doesn't expose (trigger zone polygons, per-class AI
confidence thresholds, per-preset alert-skip flags, ONVIF event handlers,
alert action definitions). It expects:
A
cam settings/directory in the launch CWD with<short>.regexports, ORBI_MCP_REG_DIRpointing at the directory.
Parsing is in-process. python-registry ships as a normal bi-mcp
dependency, so the install is single-step.
The BI_MCP_REG_DIR default resolves relative to the current working
directory at call time, not the install location, so it works correctly
whether bi-mcp is run from an editable checkout, a wheel, or via uvx.
Camera short names are validated ([A-Za-z0-9_-]+) before being composed
into a path, so a malformed name can't escape the configured directory.
Re-export a camera from BI any time you tune settings (right-click camera → Camera settings → Copy/import → Export). Files older than 7 days trigger a staleness warning in the tool's response.
Troubleshooting
unreachable: cannot reach Blue Iris.
Check
BI_HOSTandBI_PORTin.env(default port is 81).Confirm Blue Iris's web server is enabled: BI → Settings → Web server.
From the same machine, try
curl -X POST -d '{"cmd":"login"}' http://HOST:PORT/json. You should get a JSON response.
auth: Blue Iris rejected the login.
Check
BI_USERandBI_PASS.The user must have LAN access enabled in BI → Settings → Users.
Don't keep retrying with wrong credentials. Blue Iris will lock the account.
not_found: requested camera, clip, or alert doesn't exist.
For
bi_get_camera_config short=…, the value must match a camera's short name, not its display name. Runbi_list_camerasto see the list.
Empty or weird responses: pass raw=true to see what BI actually returned, then file an issue with the raw JSON so the shaper can be improved.
Debug logging: set BI_MCP_DEBUG=1 in .env (or in the MCP config env).
Logs go to stderr and a rotating file under your platform's user-cache dir:
Linux:
~/.cache/bi-mcp/server.logmacOS:
~/Library/Caches/bi-mcp/server.logWindows:
%LOCALAPPDATA%\bi-mcp\server.log
Security
This server is designed to run locally, talking to a Blue Iris box on the same LAN. It does not listen on a network port; it speaks stdio to one MCP client at a time. It should not be exposed to the internet.
Create a dedicated low-privilege Blue Iris user for the read tools (BI → Settings → Users → +):
Setting | Value | Why |
Access | Local + LAN | Required for the server to authenticate. |
Admin | unchecked | Read tools don't need it; admin-gated reads use a separate user (below). |
Change profile | unchecked |
|
PTZ | checked | Needed for |
Audio | unchecked | Not used. |
Clips | checked | Needed for |
Clip create | checked only if you enable | Required by the BI |
Camera groups | (tick all you want Claude to see) |
If you work from a clone of this repo, .env is gitignored and you should
never commit it. If credentials end up in a commit by accident, delete the
Blue Iris user immediately and create a new one.
Admin-gated tools and the two-user setup
Blue Iris gates several JSON cmds behind admin, so the recommended pattern is a two-user setup:
BI_USERandBI_PASS: the low-privilege account with PTZ and Clips access, used for all read tools that don't need admin.BI_ADMIN_USERandBI_ADMIN_PASS: a dedicated admin account, used only for the admin-gated cmds (marked ✓ in the Admin? column of the tool table above). Coversbi_get_sysconfig,bi_list_log, the deep path ofbi_get_camera_config,bi_get_camera_motion_config,bi_trigger_camera,bi_set_profile,bi_export_clip, andbi_set_camera(all ops).
If only the read user is configured, admin-gated tools degrade gracefully
(deep bi_get_camera_config falls back to the shallow camlist view; other
admin-gated tools raise a clear admin_required error). See AGENTS.md
§ Mutation patterns before enabling the mutating tools.
License
MIT. See LICENSE.
Contributions welcome. Open an issue on GitHub before sending a large PR so we can agree on shape.
Available Tools
18 toolsbi_audit_actionsARead-only
Informational tool — surfaces cross-camera action-row outliers for user review. Walks every camera's .reg export, buckets action rows into cohorts by (type, description, type-specific key), and reports fields where one camera's value deviates from the cohort's modal value under 'outliers'; fields with no strict majority appear there as entries with no_majority=true listing each value and its members. Per-camera path tokens (e.g. 'ai/SecCam_3/motion') are templated to '' before comparison so legitimate per-camera substitution doesn't false-positive. The 'enabled' field is reported separately under 'disabled_outliers' so a row left disabled by accident is easy to spot. Outliers are NOT necessarily bugs — they may be intentional per-camera customizations (e.g. one camera filtering different trigger sources, or running a narrower profile set). Present findings to the user as 'values worth confirming' and ask whether each is intentional. Pure read; no live BI connection.
Inputs: raw, hook, cameras, min_cohort.
bi_explain_alert_chainARead-only
Use after bi_list_alerts to decode what actions fired on a specific alert. Pass the alert's path from that response. Explain a specific alert's action chain. Given (camera, alert_path), returns the alert's facts (memo, profile/preset at trigger, zones), each action row with its decoded filters, comparator verdicts for the cases that need them (compound predicates like 'car+licenseplate', confidence thresholds like 'person:80', cross-zone sequencing, wait-row gating), and a ±2-minute log cross-reference of what BI actually did (MQTT publishes, email/SMS/FTP results, AI cancellations). Simple filter matches (object-in-list, profile, source bit) are surfaced as raw facts; the caller decides FIRED vs SUPPRESSED for those. Admin-gated (uses the log cmd).
Inputs: path (required), camera (required), raw, max_alert_age_h.
bi_get_actionsetARead-only
Return the semantic action set (OnTrigger and/or OnReset) for a camera. Decodes the full action type map (0-13), the command table for type=12 do-commands (PTZ presets 2201-2456, action sets, brightness/contrast/gain, plus ~60 individual codes), web_proto1 (http/https/mqtt), run_action, trig_allzones, and the profiles/zones/diobits/trig_source bitmasks into readable lists. Unmapped values fall through with the raw int preserved alongside (e.g. command_raw, trig_source_raw). Source data comes from the camera's .reg export, so changes made via the BI UI mid-session won't be visible until a re-export.
Inputs: camera (required), raw, hook.
bi_get_alert_imageARead-only
Fetch the STORED alert image (the frame BI saved when an alert fired) for a camera, resolved by time — not a live frame (that's bi_get_camera_snapshot). Pass 'camera' (short name) and optional 'at' (the alert time: ISO-8601, unix epoch int, or relative like '-2h'); omit 'at' for the most recent alert. Returns the most-recent alert at-or-before 'at' as base64, plus its record/time/memo so you can confirm which alert came back. Optional 'markup' (bool) requests the AI-overlay variant (manual's v=2). Internally resolves via alertlist + the /alerts/@record endpoint. Use a specific camera, not 'Index'. MARKUP — when 'markup'=true draws no box, it's almost always the ALERT'S SOURCE, not a tool bug: a box exists ONLY if CodeProject.AI classified the alert (memo has a score, e.g. person:89%). A bare person memo (no %) is an ONVIF/camera-IVS alert that BI's AI never scored, so there is no box to burn — v=2 just re-encodes the frame. (Separately, low-res thumbnails come from the camera's Hi-res-JPEG alert setting being off; that's a distinct issue from markup.) The /alerts/ endpoint only serves what BI stored and ignores w/h/scale params. State the source-vs-bug distinction proactively.
Inputs: camera (required), at, raw, markup.
bi_get_camera_configARead-only
Per-camera config + state. With admin creds, calls camconfig to return motion sensitivity, AI zones, recording mode, stream paths, schedule/profile flags. Without admin, falls back to filtered camlist state. Trigger zone polygons, per-class AI thresholds, and alert action definitions are NOT exposed by BI's JSON API — use bi_get_reg for those.
Inputs: short (required), raw.
bi_get_camera_motion_configARead-only
Live motion + post-trigger settings for a camera, read from BI's admin camconfig cmd. Use this instead of bi_get_reg(key_path='Motion') to avoid stale .reg exports when tuning sensitivity/contrast/breaktime. Returns motion (12 keys: sense, contrast, breaktime, maketime, usemask, objects, ai_zones, shadows, luminance, showmotion, audio_trigger, audio_sense) and post (timed, timed_interval) plus verbatim motion_raw / post_raw twins. AI thresholds (smartconf, smartlabels, periodic, static-objects) are NOT in camconfig — use bi_get_reg(key_path='AI\') for those. Trigger-zone polygons stay in bi_get_reg(key_path='Motion') under maskbits_*. Admin-required. Note: the camconfig set-half for setmotion/setpost is a silent no-op in 5.9.9.71 — this tool is read-only by design; tune in the BI UI.
Inputs: short (required), raw.
bi_get_camera_snapshotARead-only
Fetch a single current JPEG frame from a camera via GET /image/<short>. Returns the image as base64 — the calling agent decides where (if anywhere) to write it to disk. Useful for cross-referencing live camera coverage against spatial maps, verifying PTZ preset framing, or capturing a still without going through the alert/clip pipeline.
Inputs: camera (required), raw.
bi_get_clip_infoARead-only
Forensic detail for one clip/alert: resolution, duration, AI/profile/schedule/zones active at trigger time. Pass clip 'path' from bi_list_alerts.
Inputs: path (required), raw.
bi_get_ptz_statusARead-only
PTZ state for one camera. All fields BI returns are passed through (presets[], presetnum, brightness, contrast, irmode, powermode, talksamplerate). Adds two derived helpers: 'preset_map' = {N: description, ...} keyed by preset number (UI3 source: presets[] is 1-indexed by position; "(undefined)" and empty descriptions are dropped), and 'active_preset' = {num, description} when presetnum is set. Camera must have PTZ enabled in BI.
Inputs: camera (required), raw.
bi_get_regARead-only
Parse a camera's .reg export and return the requested key subtree. Use this for what the BI JSON API does NOT expose: trigger zone polygons (Motion<profile>\maskbits_*), per-class AI confidence thresholds (AI<profile>\smartconf), per-preset alert-skip flags (PTZ\Presets<n>\noalerts), ONVIF event handlers (camevents<n>), and alert action definitions (Alerts\OnTrigger). Optional 'key_path' limits the response to that subtree (e.g. 'AI\3' for profile 3 AI config). Returns staleness warning if the .reg file is >7 days old.
Inputs: camera (required), raw, key_path, include_masks.
bi_get_sessionARead-only
Blue Iris version/license, time zone, capabilities of the current user (admin/ptz/clips/etc), and available profile/schedule/stream names.
Inputs: raw.
bi_get_statusBRead-only
Snapshot of Blue Iris system state: active profile, schedule hold/run, CPU%, RAM, disk usage, uptime, DIO outputs, warnings.
Inputs: raw.
bi_get_sysconfigARead-only
System config snapshot (admin required): FTP archive enable, global schedule on/off, manual record time limit, plus any DIO/MQTT state BI exposes inline. Use this instead of asking the user to screenshot Settings → Other.
Inputs: raw.
bi_get_timelineARead-only
24-hour activity timeline (motion/trigger/alert buckets) for a camera. Requires 'camera' short name. If both 'startdate' and 'enddate' are omitted, defaults to the last 24 hours (BI returns empty spans for a rangeless query).
Inputs: camera (required), raw, msecpp, enddate, +1 more.
bi_list_alertsARead-only
Recent alerts with AI memo (object, confidence, license plate), zones triggered, and clip path. Requires 'camera' short name (or 'Index' for all). First stop for 'what fired when' / reconstructing an alert chain — per-alert timestamps and memos, no dedup. Use this before bi_list_log when investigating a specific event. Optional 'startdate'/'enddate' accept unix epoch int, ISO-8601 ('2026-05-27T14:09:02Z'), or relative shorthand ('-2h', '-1d'). 'view' (filter; see schema for full enum), 'search' (memo substring). 'limit' default 50. Crossover note: if 'view' is set to 'flagged', BI may also return clip items here; those clips lack the 'zones' field and their 'msec' is the clip length, not alert length.
Inputs: camera (required), raw, view, limit, +3 more.
bi_list_camerasBRead-only
List of all cameras and groups: online state, motion/trigger/alert counts, stream bitrate/FPS/resolution, last alert time, error state.
Inputs: raw, limit.
bi_list_clipsARead-only
Recent recorded clips for a camera: path, duration, resolution, flags, memo. Complementary to bi_list_alerts (clips include continuous recordings; alerts are AI/motion events). Requires 'camera' short name (or 'Index' for all). Optional 'view' (filter; see schema for full enum), 'startdate'/'enddate' accept unix epoch int, ISO-8601, or relative shorthand ('-2h', '-1d'). 'search' (memo substring, server-side), 'tiles' (true=one entry per day, useful for calendar views). 'limit' default 50. Crossover note: alert-side view values (e.g. 'alerts', 'people', 'zonea') will return alert items in this response — they have an 'msec' field meaning alert length (not clip length) and lack the 'zones' field. UI3 v91 fixed a bug where this was mishandled.
Inputs: camera (required), raw, view, limit, +4 more.
bi_list_logARead-only
Recent Blue Iris system log entries with optional filters.
Pick the right tool: for reconstructing 'what fired when' on a camera, start with bi_list_alerts — per-alert timestamps with no dedup. This log is best for system events (profile changes, disk ops, logins, errors) and aggregate activity counts.
Filters:
since — UTC epoch sec, ISO-8601, or '-15m'/'-2h'/'-1d' (server-side via aftertime)
camera — exact match on entry.obj (clone cameras log under their own short names)
obj — exact match on entry.obj (escape hatch: 'App', 'MQTT', 'DB', 'AI_Input', drive letters, usernames)
levels — list of accepted level ints; empirical: 0=info, 1=warn, 2=error, 3=trigger/alert aggregate (deduped — use bi_list_alerts for per-event), 4=status change, 10=user
match — case-insensitive substring on entry.msg
regex — Python regex on entry.msg (IGNORECASE); xor with match
limit — applied AFTER filtering (default 100)
Returns {entries, scanned, matched, warning?}. raw=true bypasses the envelope and shaper. Admin required.
BI aggregates repeated messages: count is cumulative since BI startup (or last log clear), and date is when BI last summed the entry, not necessarily the most recent occurrence. To tell whether a message is actively firing now, re-query with a tight since=-5m window.
Inputs: obj, raw, limit, match, +4 more.
Tool Schema Changelog
Recent tool additions, removals, and schema changes observed during successful MCP inspections.
Loading tool schema changelog…
TDQS
Scored across 18 tools
Most tools target clearly distinct resources (snapshot vs alert image, alerts vs clips vs log, camera list vs camera config). The config surface is crowded — bi_get_camera_config, bi_get_camera_motion_config, bi_get_actionset, bi_get_reg, and bi_audit_actions all surface overlapping config data — but the descriptions explicitly steer between them (e.g. 'use this instead of bi_get_reg(key_path=...)' and the API-vs-.reg boundary). Crossover notes on alerts/clips/log further reduce misselection.
Every tool uses the consistent bi_<verb>_<noun> pattern with clear verbs (get, list, audit, explain) and snake_case nouns throughout. The bi_ prefix is applied uniformly, and verb choice signals read semantics reliably.
18 tools is at the upper edge but justified by a genuinely broad domain (cameras, alerts, clips, logs, config, PTZ, timeline, audit). Each tool maps to a distinct facet rather than being redundant padding, though the five-way config split could be trimmed.
The read/inspection surface is thorough: status, session, sysconfig, cameras, per-camera config, motion config, actions, reg, alerts, clips, log, timeline, images, and clip forensics cover the analyst workflow well. The main gap is the absence of any mutation operations (profile/schedule changes, PTZ moves, clip export), which the descriptions acknowledge as read-only by design.
Maintenance
Related MCP Connectors
Ask your security cameras anything and set up alert rules in a sentence. Built into Agent DVR.
Watch cameras and video with AI vision and get alerts in Slack, Discord, Telegram or a webhook.
Connect Claude to Fathom meeting recordings, transcripts, and summaries
Connect ChatGPT or Claude to your CRM, store or shipping API. Keys stay out of the chat.
Related MCP Servers
- -licenseAqualityNot gradedmaintenanceA bridge between AI assistants like Claude and Anava-enabled Axis cameras, enabling real-time image analysis, event monitoring, and camera management through natural language commands.4-
- AlicenseNot gradedqualityDmaintenanceConnects AI assistants like Claude to Jira projects, enabling natural language queries and operations for issue management, project tracking, comments, and workflows through the Jira REST API.21,328 npm79ISC
- FlicenseNot gradedqualityDmaintenanceConnects Claude to OpenNMS, allowing plain language interaction with alarms, nodes, events, asset records, categories, and service collection.1-
- FlicenseAqualityBmaintenanceEnables AI assistants to interact with Frigate NVR security camera systems, supporting camera management, event detection, snapshots, recordings, and system stats via natural language.714-