Eos Fitting MCP Server
MCP server that wraps the Eos EVE Online fitting engine. Agents propose fit changes; this server applies them, recalculates, and returns a FitReport.
README
Pyfa MCP
MCP server that wraps the Eos EVE Online fitting engine (same stack as Pyfa). Agents propose fit changes; this server applies them, recalculates, and returns a FitReport. It never suggests modules — it only evaluates.
Install (uvx)
Once published to PyPI:
uvx pyfa-mcp
Until then, from a checkout (submodules required so Eos is bundled into the install):
git clone --recurse-submodules git@github.com:theonlysinjin/eve-fit-mcp.git
cd eve-fit-mcp
uvx --from . pyfa-mcp
Or from git directly (after submodules are fetchable in the build):
uvx --from git+https://github.com/theonlysinjin/eve-fit-mcp.git pyfa-mcp
Cursor MCP config (uvx)
{
"mcpServers": {
"pyfa-mcp": {
"command": "uvx",
"args": ["--from", "/path/to/eve-fit-mcp", "pyfa-mcp"]
}
}
}
After PyPI:
{
"mcpServers": {
"pyfa-mcp": {
"command": "uvx",
"args": ["pyfa-mcp"]
}
}
}
Staticdata downloads on first run into ~/.cache/pyfa-mcp/ (or call refresh_static_data). First Eos cache build can take a few minutes.
Requirements
- Python 3.10+ (pulled in by uvx)
- Network on first run for staticdata (or set
EOS_PHOBOS_PATH) - Dev checkouts: git submodules
eos,phobos,pyfa(pyfa/staticdataused when present)
Staticdata (auto / refresh)
EOS_PHOBOS_PATHif set- In-tree
pyfa/staticdata(submodule) ~/.cache/pyfa-mcp/staticdata(orEOS_DATA_DIR)- Download from GitHub release
staticdata
Environment
| Variable | Required | Description |
|---|---|---|
EOS_PHOBOS_PATH |
no | Dump root; auto-resolved if omitted |
EOS_CACHE_PATH |
no | Eos cache file (default under data dir) |
EOS_DATA_DIR |
no | Download/cache root (default ~/.cache/pyfa-mcp) |
EOS_STATICDATA_URL |
no | Override release asset URL |
EOS_SOURCE_ALIAS |
no | Default tq |
EOS_PACKAGE_PATH |
no | Only needed for editable/dev without wheel-bundled eos |
EOS_MAX_FITS |
no | Max in-memory fits (default 100) |
EOS_FIT_TTL |
no | Optional fit TTL in seconds |
Dev install
git clone --recurse-submodules git@github.com:theonlysinjin/eve-fit-mcp.git
cd eve-fit-mcp
uv venv .venv && source .venv/bin/activate
uv pip install -e ".[dev]"
pytest
pyfa-mcp # or: python -m pyfa_mcp
Prebuilt binary
See binary-latest (pyfa-mcp-linux-x64, macOS, Windows). Point Cursor at the binary; staticdata still auto-downloads.
Creating AGENTS.md (fitting projects)
Put an AGENTS.md in the project where you design fits (not in this MCP repo):
# Fit with Pyfa MCP
You design EVE Online fits by proposing changes; the **pyfa-mcp** MCP evaluates them. It never suggests modules — you do.
Use the **eve-online-esi** MCP when the fit should reflect a real character’s skills. Prefer a cached map under `users/<Name>/skills.json`; refresh from ESI when asked or when the file is missing/stale.
## Skills (player maps)
**Apply (fitting):** load `users/<Name>/skills.json` → `set_skills(fit_id, data["skills"])`. Keys are skill type IDs; values are `active_skill_level` (0–5). Theorycraft: skip and use `apply_all_skills_5`.
**Refresh (update cache):**
1. `add_character` if needed (SSO; tokens stay local). Re-auth on 401 / missing character.
2. `GetCharactersCharacterIdSkills` with `character_id` (+ `X-Compatibility-Date`).
3. Write `users/<Name>/skills.json` as `{ character_id, name, updated_at, total_sp?, unallocated_sp?, skills: { str(skill_id): active_skill_level } }`.
## Startup
1. Confirm the goal in one line: role, constraints (EHP, DPS, tank type, cap stable?), and skills (player map via `users/…/skills.json`, or theorycraft).
2. If using a player map: load (or refresh) skills as above.
3. `create_fit(ship_type_id)` then either `set_skills` (from file) or `apply_all_skills_5`.
4. Rough in a full fit with type IDs: highs → mids → lows → rigs → drones/fighters → implants if needed.
5. Read the FitReport. Fix hard blockers first (`validation_errors`, CPU/PG/slots), then optimize toward the goal.
6. Iterate: **one** change per turn (`equip_module` / `replace_module` / `set_module_state` / `set_charge` / …). Compare reports. Use `clone_fit` for A/B forks.
## Rules
- Type IDs only — never invent them.
- Soft failures (CPU, skills, slots) still apply; hard errors (bad ID, wrong rack) do not mutate.
- Racks: `high` / `mid` / `low`. States: `offline` | `online` | `active` | `overload`.
- Stop when constraints are met or gains flatten. Summarize the final fit + key stats.
## FitReport priorities
`validation_errors` → `resources` / `slots` → `combat` (dps, ehp, RPS) → `mobility` → `fit` snapshot.
Tools (v1)
Session: create_fit, clone_fit, delete_fit, list_fits, get_fit, reset_fit
Skills: set_skills, set_skill, clear_skills, apply_all_skills_5
Hull: set_ship, set_stance, equip_module, replace_module, remove_module, set_module_state, set_charge, add_rig / remove_rig, add_subsystem / remove_subsystem, add_drone / remove_drone / set_drone_state, add_fighter / remove_fighter / set_fighter_state, add_implant / remove_implant, add_booster / remove_booster, set_effect_beacon
Eval: get_stats, validate_fit
Data: refresh_static_data
Non-goals
- Autofitting / “make this better”
- ESI login, skill sync, market prices
- wx/GUI / Pyfa desktop integration
- Full EFT/DNA import in v1
Recommended Servers
playwright-mcp
A Model Context Protocol server that enables LLMs to interact with web pages through structured accessibility snapshots without requiring vision models or screenshots.
Audiense Insights MCP Server
Enables interaction with Audiense Insights accounts via the Model Context Protocol, facilitating the extraction and analysis of marketing insights and audience data including demographics, behavior, and influencer engagement.
Magic Component Platform (MCP)
An AI-powered tool that generates modern UI components from natural language descriptions, integrating with popular IDEs to streamline UI development workflow.
VeyraX MCP
Single MCP tool to connect all your favorite tools: Gmail, Calendar and 40 more.
graphlit-mcp-server
The Model Context Protocol (MCP) Server enables integration between MCP clients and the Graphlit service. Ingest anything from Slack to Gmail to podcast feeds, in addition to web crawling, into a Graphlit project - and then retrieve relevant contents from the MCP client.
Kagi MCP Server
An MCP server that integrates Kagi search capabilities with Claude AI, enabling Claude to perform real-time web searches when answering questions that require up-to-date information.
E2B
Using MCP to run code via e2b.
Neon Database
MCP server for interacting with Neon Management API and databases
Exa Search
A Model Context Protocol (MCP) server lets AI assistants like Claude use the Exa AI Search API for web searches. This setup allows AI models to get real-time web information in a safe and controlled way.
Qdrant Server
This repository is an example of how to create a MCP server for Qdrant, a vector search engine.