Skip to main content
Glama
ED Corai

ED Corai

by kay-dutch

ED-Corai

Ship builds for Elite Dangerous worked out by the numbers, not by forum consensus.

Licence Release CI MCP Rust Node.js Elite Dangerous

ED Corai is an MCP server: it plugs into Claude, and into any other AI app that supports the Model Context Protocol (OpenAI Codex, Gemini CLI, GitHub Copilot in VS Code, Cursor and others).

Ask for a build in plain words — "longest-range Krait Mk II with a fuel scoop, AFMU and SRV bay", "tankiest Fer-de-Lance that still does 150 sustained DPS", "which medium ship carries the most cargo and still jumps 20 Ly laden?" — and ED Corai searches the outfitting space, scoring every candidate with the full outfitting math: mass, power, jump range, speed, shields and hull with resistances, sustained DPS. You get the build, the numbers behind it, and links to open it in Coriolis or EDSY.

What it does

  • Finds the best fit for one objective under hard constraints (minimum jump range, required modules, cost cap, power budget, weapon consistency, ...).

  • Engineering included by default: grade-5 blueprints at best roll plus the best experimental effect for each module. Can be switched off for stock builds.

  • Compares hulls: runs the same goal over every ship (or a subset) and ranks them.

  • Evaluates and tweaks existing builds: paste a Coriolis link, an EDSY export, or a journal Loadout, change a slot, see the effect.

  • Converts builds between formats: paste a Coriolis link, an EDSY SLEF export or a journal Loadout and get it back as a Coriolis link, an EDSY import link, SLEF or journal JSON (for EDSY, Coriolis, Inara and EDDiscovery).

  • Reports constraints it could not meet instead of quietly ignoring them.

Tools exposed to the assistant

Tool

Purpose

list_ships

Hulls with size, price, base stats and slot layout

ship_details

Slots of a ship and its stock build

list_module_groups

Module group codes (fs fuel scoop, sg shield generator, ...)

module_info

Module stats and the engineering available for a group

list_metrics

Everything that can be optimised or constrained

evaluate_build

Full stats of a build, optionally after slot changes

export_build

A build as SLEF or journal Loadout JSON, plus Coriolis and EDSY links

optimize_build

Best outfitting of one ship for an objective and constraints

compare_ships

Same optimisation across many hulls, ranked

current_ship

The ship you fly now, read from the game journal, fully evaluated

recent_ships

Your recently flown ships with key figures and links

check_for_updates

Installed version, the game data it was built from, and the latest release

install_update

Updates the install to a release, after you agree

Builds can be given in any common form: a Coriolis outfit or import link, an EDSY import link, SLEF text (in EDSY: Export → SLEF), or a journal Loadout event. EDSY's short #/L= links cannot be read; use its SLEF export instead.

Metrics

Travel: jump ranges (unladen, laden, full tank, total per tank), speed, boost, pitch/yaw/roll, boost interval. Capacity: cargo, passengers, fuel, fuel scoop rate, refinery bins, collector limpets, dry and laden mass, cost, power margin.

Defence: raw and effective shield HP against a damage mix, worst-type shield HP, shield regen, recharge and recovery times, time shields last under a given incoming DPS (with or without cell banks), raw and effective hull HP, module protection.

Offence: sustained and burst DPS, DPS against a reference target's shields and hull, piercing efficiency against its hardness, time to kill it, weapon capacitor sustain, ammo endurance, weapon heat, number of weapon types and mount types (for consistent, aimable loadouts).

Heat: the heat levels coriolis.io shows (idle, full thrust, FSD charging, all weapons firing with the capacitor full or nearly empty, shield cells spinning up) and the time until the ship overheats while firing, from the hull's heat capacity and dissipation; plus net heat per second in a fight with Thermal Vent beams and the vent strength at which the ship would hold steady.

Plus count:<group> / class:<group> for "must have" modules.

Related MCP server: eve-fit

How the numbers are computed

Everything is calculated, nothing is looked up from community tier lists. Where the game's own mechanics are known they are applied directly; where tools disagree or no measurement exists, the assumption is a setting you can change.

  • Resistances follow the game's stacking rules: shield booster resistances get diminishing returns above 30 %, the generator's own resistance is exempt, SYS pips add 60 % × (pips/4)^0.85 on shields only. Hull reinforcement resistances stack one module at a time, weakest first: a module's resistance above 30 % counts double, the combined value gets diminishing returns between 30 % and 65 %, and the total caps at 75 %.

  • Hardness vs piercing reduces hull damage by piercing / hardness when a weapon's penetration is below the target's hardness; this is why piercing efficiency and time to kill are reported against a reference opponent.

  • Jump range counts fuel once and excludes the reserve tank by default, matching the journal's MaxJumpRange (Coriolis counts the reserve tank; reserveFuelInJumpMass restores its behaviour).

  • Shields under fire: regeneration is assumed to pause while taking hits (regenUnderFire = 0). The real pause is unmeasured; set 0–1 to explore.

  • Corrosive Shell raises all hull damage to a target it keeps hitting by 25 % (once, however many weapons carry it) and adds 20 piercing to its own weapon. Coriolis models only its ammo penalty. Both figures are community numbers and can be changed.

  • Thermal Vent beams generate no heat while on target and remove heat instead. How much they remove is unmeasured; the default assumes as much as the beam would otherwise generate.

  • Ammo is finite: ammoEndurance is the time until the first ammo-limited weapon runs dry, so missiles and torpedoes are not mistaken for sustained damage. Plasma Slug counts as unlimited.

  • Outfitting rules mirror the game: life support, sensors and SCO frame shift drives must match their slot size; thrusters must be able to lift the hull; one shield generator, fuel scoop, refinery, interdictor, scanner of each kind, and so on; at most four experimental (AX/Guardian) weapons; restricted slots (military, cargo, limpets, mining hardpoints) and ship-specific modules are honoured.

  • Heat follows Coriolis' model (from Frontier forum research): heat settles at √(load ÷ dissipation) up to the hull's maximum dissipation, then climbs until the ship overheats; weapons heat up to five times more as the capacitor runs dry. The Anaconda Mk II borrows the old Anaconda's dissipation until Coriolis has its own; the Type-7 has no heat figures yet.

  • Not modelled yet: sensor range, module breach chance, NPC behaviour. The default pve damage mix (30 % kinetic, 60 % thermal, 5 % explosive, 5 % absolute) is an assumption, not a measurement.

Every default that rests on an assumption is listed by list_metrics so you can rerun with the alternative when a ranking is close.

Installation

Download the zip for your system from the latest release and unzip it into a folder that will stay where it is (for example Documents/ed-corai):

  • Windows: ed-corai-<version>-windows-x64.zip

  • macOS (Apple silicon and Intel): ed-corai-<version>-macos-universal.zip

Inside are two programs; neither needs anything else installed:

  • ed-corai-cli (ed-corai-cli.exe on Windows): the MCP server your AI app starts. Note its full path: every app below asks for it.

  • ed-corai (ed-corai.exe): the optional desktop window.

The programs are not code-signed yet, so the system distrusts them once:

  • Windows: before unzipping, right-click the zip → Properties → tick Unblock → OK.

  • macOS: after unzipping, run once in Terminal: xattr -dr com.apple.quarantine /path/to/ed-corai

Check that it runs: /path/to/ed-corai/ed-corai-cli version.

From source

With Rust 1.88 or newer: git clone --recursive https://github.com/kay-dutch/ed-corai.git, then cargo build --release -p ed-corai-cli; the program is target/release/ed-corai-cli.

The original Node.js version is in the same repository (Node.js 20 or newer): npm install, npm run build, then use the command node with the argument /path/to/ed-corai/dist/server.cjs wherever the instructions below say ed-corai-cli and mcp. It updates itself with git instead of downloads.

Configuration

Every app needs the same two things: the full path to ed-corai-cli as the command, and mcp as its argument. In the examples, replace PATH with yours, e.g. C:\\Users\\you\\Documents\\ed-corai\\ed-corai-cli.exe on Windows (JSON needs doubled backslashes) or /Users/you/Documents/ed-corai/ed-corai-cli on macOS.

Claude Code

claude mcp add ed-corai -s user -- PATH mcp

-s user makes it available in every project; without it, it only works in the folder where you ran the command.

Claude Desktop

Add to claude_desktop_config.json (Settings → Developer → Edit Config):

{
  "mcpServers": {
    "ed-corai": {
      "command": "PATH",
      "args": ["mcp"]
    }
  }
}

Restart Claude Desktop afterwards.

Other MCP apps

ED Corai runs on your computer, so it works in any app that can start a local MCP server.

OpenAI Codex CLI, in ~/.codex/config.toml:

[mcp_servers.ed-corai]
command = "PATH"
args = ["mcp"]

Gemini CLI, in ~/.gemini/settings.json:

{
  "mcpServers": {
    "ed-corai": {
      "command": "PATH",
      "args": ["mcp"]
    }
  }
}

GitHub Copilot in VS Code (agent mode), in .vscode/mcp.json, or run MCP: Add Server from the command palette:

{
  "servers": {
    "ed-corai": {
      "type": "stdio",
      "command": "PATH",
      "args": ["mcp"]
    }
  }
}

Cursor, Windsurf, Cline, LM Studio and most other apps take the same mcpServers block as Claude Desktop above, in their own MCP settings file.

Your own agent: with the OpenAI Agents SDK, start it as a stdio server:

from agents.mcp import MCPServerStdio

ed_corai = MCPServerStdio(params={"command": "PATH", "args": ["mcp"]})

Things to know outside Claude:

  • Use a large model. A good answer takes several tool calls and reads large results. Big models (GPT-5, Gemini Pro, Claude) handle it; small local models often lose the thread.

  • Some apps drop the server's instructions. ED Corai sends a short guide on how to use its tools (lock the slots you are not changing, keep what the build already does as minimums). If results look careless, tell the model: "Read list_metrics first. Lock every slot I don't mention."

  • Ask for updates yourself. Not every app shows the model the update notice; ask "check for ED Corai updates" now and then.

  • ChatGPT (app and website) only connects to MCP servers hosted on the internet, not to programs on your computer, so it cannot use ED Corai as it is.

Options

Most behaviour is chosen per request, and the assistant picks sensible values from your wording. You can also ask for them directly:

Option

Default

Meaning

engineering

full

full = G5 + experimentals, none = stock modules

effort

normal

quick, normal or thorough (more restarts, slower)

power

priority

priority: combat modules stay powered with hardpoints out while non-combat ones (scoop, docking computer, SRV bay, ...) are shed by priority group, as in game; deployed: everything on with hardpoints out; retracted; ignore

requireBoost

true

Distributor must be able to boost

lock

—

Slots to keep as they are, e.g. ["hardpoint", "standard:fsd"]

filters

everything allowed

Forbid Powerplay, pre-engineered, community-goal, Guardian or AX modules, or specific groups

context.sysPips / wepPips

2 / 4

Pips while fighting: shield resistance, regen under fire, weapon sustain

context.recoverySysPips

4

SYS pips while recovering: shield recharge and rebuild times

context.damagePreset

even

even or pve (thermal-heavy); explicit damage weights override

context.incomingDps

200

Raw incoming DPS for time-under-fire metrics

context.regenUnderFire

0

Share of time shields regenerate while under fire

context.opponent / rangeMeters

stock Anaconda / 1000

Reference target and range for offence metrics

context.ventHitRate / thermalVentFactor

1 / 1

Share of time Thermal Vent beams hit, and heat removed per hit relative to the beam's own heat

context.corrosiveHullMultiplier / corrosivePiercingBonus

1.25 / 20

Corrosive Shell hull damage multiplier and piercing bonus

context.reserveFuelInJumpMass

false

Count the reserve tank in jump mass (Coriolis behaviour)

Environment variables for the server:

Variable

Meaning

CORAI_DEBUG=1

Log diagnostic output to stderr

CORAI_UPDATE_CHECK=0

Do not look for new releases at start-up

CORAI_REPO=owner/name

Check a fork for releases instead of this repository

CORAI_JOURNAL_DIR=path

Read the game journal from another folder

Saved builds (SLEF, Loadout JSON) can be passed to the tools by path on the command line, e.g. ed-corai-cli evaluate builds/my-corvette.slef or ed-corai-cli optimize builds/my-corvette.slef jumpLaden --lock hardpoint.

Desktop window

Start ed-corai (ed-corai.exe) to watch ED Corai work. It is optional: your AI app uses ED Corai the same with or without it.

  • The station turns while a tool runs; the window shows the request, the AI app that sent it and the results, each with a link to open the build in Coriolis.

  • It lists the AI apps connected to ED Corai and the ship you fly now, from the game journal.

  • Defaults for your AI app: how you fly (SYS and WEP pips in combat, NPC or even damage, stock or engineered modules). ED Corai uses them whenever a request does not say otherwise, and tells the assistant so.

  • It offers new releases and installs them for both programs.

Pin it to the taskbar or Dock if you like to keep it open next to the game.

Updating

At start-up ED Corai checks GitHub for a newer release, at most once per launch, and the assistant tells you when one is out. Say yes and it updates itself: it downloads the release for your system, checks it against the release's published checksum and replaces the program. Then restart it (/mcp in Claude Code, or restart your app). You can also ask "is there an update?" at any time, run ed-corai-cli update yourself, or install it from the desktop window.

By hand: download the new zip and unzip it over the old folder (on macOS, repeat the xattr step).

Each release states which Coriolis data it was built from; check_for_updates shows it for your install. New ships and modules reach ED Corai once they are in Coriolis.

Using it

Talk to the assistant as you would to an experienced builder. It turns the request into an objective and constraints, runs the tools and answers with the numbers and links.

Find a build

Longest-range Krait Mk II with a fuel scoop, AFMU and SRV bay.

Tankiest Fer-de-Lance that still does 150 sustained DPS. Fixed weapons only, one weapon type.

Mining Python with the most cargo that keeps 20 Ly laden range.

Pick a ship

Which medium ship gives the most shield for PvE combat zones? Show the top five.

I have 400 million credits. Best laden jump range for a trader?

Improve your own build

Paste a Coriolis link, an EDSY SLEF export or a journal Loadout.

Here is my Anaconda Mk II: . Keep the weapons and FSD. Make it run cooler without losing shields or DPS.

This is my Corvette . Get the Anaconda as close to it as possible in protection and damage.

Your ships in the game

ED Corai reads the game journal on your PC (Saved Games\Frontier Developments\Elite Dangerous), so the assistant can see what you fly.

Look at my current ship. What would make it jump further without losing shields?

Which of my recent ships is best for a massacre mission?

What if

What happens if I swap the multi-cannon on hardpoint 6 for a Thermal Vent beam?

Same build, but I fight with 4 pips in SYS. Does the shield choice change?

Take it to the game

Give me SLEF for EDSY and the Coriolis link.

Open this Coriolis build in EDSY:

Here is my SLEF from EDSY. Give me a Coriolis link.

Tips

  • Say what must stay. Slots you name as fixed are left alone. Other modules of your build are kept unless changing them improves what you asked for, and the answer lists every change.

  • Say what must be there: fighter hangar, docking computer, heat sink, a real fuel tank.

  • Say how you fly: pips, typical range, PvE or PvP, what you usually fight.

  • Ask for alternatives. "What does 10 % more shield cost?" shows the trade-off better than a single answer.

  • Ask which assumptions a result depends on, and rerun with the other value when it is close.

Notes

  • Results come with power priorities set (combat modules in group 1, non-combat in group 5) and internal modules moved to the smallest slots they fit; neither changes any number.

  • Results come from a heuristic search (multi-start local search with engineering, relocation and repair moves). It finds very strong builds but does not prove global optimality; thorough effort explores more.

  • Engineering is modelled at best roll. Real rolls vary.

  • Elite Dangerous is a trademark of Frontier Developments plc. ED Corai is a fan project and is not affiliated with or endorsed by Frontier.

License

MIT, see LICENSE. The bundled Coriolis engine is also MIT. Ship and module data belong to Frontier Developments plc and are not covered by this licence; see THIRD_PARTY_NOTICES.md.

Tool Schema Changelog

Recent tool additions, removals, and schema changes observed during successful MCP inspections.

Loading tool schema changelog…

    Related MCP Connectors

    Related MCP Servers

    • A
      license
      Not graded
      quality
      B
      maintenance
      Enables AI-powered Path of Exile 2 character optimization through natural language queries, providing intelligent build recommendations, gear upgrades, and passive tree optimization using the official PoE API and comprehensive game database.
      33 PyPI
      86
      MIT
    • A
      license
      A
      quality
      D
      maintenance
      Enables AI agents to search EVE Online items, validate ship fits against fitting limits, compute combat stats like DPS and EHP, and retrieve Jita market prices.
      4
      MIT
    • A
      license
      Not graded
      quality
      A
      maintenance
      Enables AI assistants to search EVE Online ships, modules and charges, build and validate fits, compute Pyfa-parity statistics, compare fits, run what-if scenarios, and rank candidate modules, charges and drones via batch optimisation. It is stateless and engine-agnostic, working with any engine implementing the eve-dogma contract over stdio or Streamable HTTP.
      MIT
    • A
      license
      A
      quality
      B
      maintenance
      Enables AI assistants to read and act on EVE Online data through the ESI API, handling SSO login for multiple characters and answering in single calls with resolved names, combined requests, and locally indexed static data. Covers fittings (with attributes and EFT/DNA export), skills and skill checks, assets, market prices and orders, industry, mail, and arbitrary ESI routes.
      35
      16 npm
      1
      MIT