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?

17 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.

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.

Heads up: the PyPI release is being prepared. Until it lands, use the git+ install command shown in the second block below. The pipx install bi-mcp form will start working as soon as v0.2.0 is on PyPI.

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 check

Until PyPI is live, install straight from the GitHub source:

# Works today, no PyPI release required
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 check

If 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 --help

If 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-server

Replace 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.json

• Windows: %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 check

A 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_3

And for poking at tool schemas interactively, the MCP Inspector opens a browser UI:

npx @modelcontextprotocol/inspector uvx --from . bi-mcp-server

Configuration 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.

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                       # or: python -m venv .venv && source .venv/bin/activate && pip install -e .[dev]
uv run bi-mcp-server check

Tool reference

17 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.

For the canonical reference with routing tables, error taxonomy, and mutation-safety rules, see AGENTS.md.

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>.reg exports, OR BI_MCP_REG_DIR pointing 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_HOST and BI_PORT in .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_USER and BI_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. Run bi_list_cameras to 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.log

• macOS: ~/Library/Caches/bi-mcp/server.log

• Windows: %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 → +):

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_USER and BI_PASS: the low-privilege account with PTZ and Clips access, used for all read tools that don't need admin.

• BI_ADMIN_USER and BI_ADMIN_PASS: a dedicated admin account, used only for the admin-gated cmds (marked ✓ in the Admin? column of the tool table above). Covers bi_get_sysconfig, bi_list_log, the deep path of bi_get_camera_config, bi_get_camera_motion_config, bi_trigger_camera, bi_set_profile, bi_export_clip, and bi_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.