compute_correlation
Compute correlation between branches
LABSCO SUMMARY
Answers the physics question instead of writing a script that answers it.
Asking for the muon pT distribution gets a computed histogram and a saved plot, which removes the write-run-debug loop that normally sits between the question and the answer. Choose the mode deliberately: core keeps the dependency set small for inspection and export, and extended is the one that brings fitting, kinematics and plotting — and it can be switched at runtime rather than at install. Before any shared deployment, set allowed_roots, because the default is deliberately permissive for local use.
What it is
An analysis layer over ROOT files. It inspects trees and branches, reads data with physics selections applied, and computes and plots the distributions — with the heavier physics tools switched on by mode rather than always loaded.
What you get
- Files browsed in a configured resource, a file's structure inspected, its integrity validated, and the branches of a TTree or RNTuple listed by pattern
- Branch data read over an entry range with a selection applied, and per-branch statistics that accept derived variables through defines
- Data exported to JSON, CSV or Parquet, with compression, straight from a selection
- 1D and 2D histograms computed with bins, range and weights, and derived variables available in the 2D case
- Histogram fitting with a model, initial guesses, bounds and fixed parameters
- Invariant mass from pt, eta and phi branches, and correlation between two branches by Pearson or Spearman
- 1D and 2D plots written to a path, either from a histogram you already computed or straight from the file
- Bin-by-bin arithmetic between two histograms — asymmetry, difference, ratio
- A mode switch at runtime between core and extended, and a call that reports which mode and capabilities are active
Requirements
pip install root-mcp, then root-mcp --data-path /path/to/your/data, or set ROOT_MCP_DATA_PATH once. The mode decides both the tools and the dependencies: core needs uproot, awkward, numpy and pandas and covers inspection, reading, statistics and export; extended adds scipy and matplotlib and brings in histograms, fitting, kinematics, correlations and plotting. Remote XRootD access needs the xrootd extra plus --allow-remote and a named resource. Native PyROOT execution is opt-in behind --enable-root with --allowed-root limiting where it may run. The permissive default, an empty allowed_roots, means any local path is reachable — it is meant for a trusted local session, and the HTTP transport refuses to start without authentication, Origin validation and a restrictive policy. The same backend also ships as a CLI. MIT.
Setup effort
One command — pip install root-mcp
USE CASES
01
Working out what is in an unfamiliar ROOT file
Inspect its structure, list the branches of the tree that matter, and validate the file before spending time on it.
02
Looking at a distribution with a cut applied
Compute the histogram over the branch with your bins, range and selection, then write the plot to a file to look at.
03
Fitting a peak
Fit the histogram with a model, supplying initial guesses, bounds or fixed parameters where the fit needs help.
04
Reconstructing a resonance
Compute invariant mass from the pt, eta and phi branches, then histogram and fit the result.
05
Comparing two samples
Compute a histogram for each and take the ratio, difference or asymmetry bin by bin, then plot what comes out.
06
Getting selected data out to another tool
Read the branches you need with the selection applied and export them as JSON, CSV or Parquet.
07
Keeping the dependency footprint small until you need more
Run in core mode for inspection and export, and switch to extended at runtime when the work turns into fitting and plotting.
17 tools
Counted once with only startup setup. Other versions, modes or optional credentials may expose a different list.
compute_histogram
Compute 1D histogram with fitting support
compute_histogram_2d
Compute 2D histogram (supports derived variables via defines)
compute_invariant_mass
Compute invariant mass from 4-vectors
export_data
Export branch data to JSON, CSV, or Parquet
fit_histogram
Fit histogram with model function
get_branch_stats
Get statistics for branches (supports derived variables via defines)
get_server_info
Get server mode and capabilities
histogram_arithmetic
Perform bin-by-bin arithmetic on two histograms (e.g. asymmetry, difference, ratio)
inspect_file
Inspect ROOT file structure and contents
list_branches
List branches in a TTree or RNTuple
list_files
List ROOT files in a resource
plot_histogram_1d
Create and save a 1D histogram plot. Provide EITHER 'data' (pre-calculated) OR 'path', 'tree_name', 'branch', 'bins' (compute from file).
plot_histogram_2d
Create and save a 2D histogram plot. Provide EITHER 'data' (pre-calculated) OR 'path', 'tree_name', 'branch_x'...' (compute from file).
read_branches
Read branch data from a TTree or RNTuple
switch_mode
Switch between core and extended modes
validate_file
Validate ROOT file integrity
BEFORE YOU INSTALL
What to know first
Tools you get
17 tools when we connected.
Where it runs
The documented client runs the server on your machine.
CONTEXT COST
Light Standing cost
Rated by the size of the 17 tool definitions that ride along on every turn, not by how many there are
INSTALLATION
Connect your client
Server setup
TERMINAL
pip install root-mcpMCP Clients
Claude Desktop
CLAUDE DESKTOP · JSON
{
"mcpServers": {
"root-mcp": {
"command": "root-mcp",
"args": [
"--data-path",
"<YOUR_PATH_TO_DATA>"
]
}
}
}
