Releases: AochiToxx/firestorm-mcp
Release list
Firestorm MCP 0.3.0a3 — public alpha
Firestorm MCP's first public alpha gives AI agents local tools for Second Life export inspection, mesh previews, viewer controls and visual checks. Built mostly by AI coding agents, under human direction.
Start here: download firestorm-mcp-0.3.0a3-source.zip, extract it to a permanent folder, then run Install.cmd on Windows or sh Install.sh on Linux/macOS. Install Python 3.11–3.14 and Firestorm separately. The installer prints the configuration to add to your agent.
- Shorter setup, capability and contributor guides, with direct download links.
- Clearer tool descriptions; runtime logic and all 43 workflow tool contracts are unchanged from 0.3.0a2.
- Optional mesh-preview skill 0.1.1 teaches the tested importer procedures.
Verified: 85 tests across eight CI jobs, fresh source/wheel installs, MCP Inspector, skill format/examples, an offline consumer review and archive checksum checks. CI results.
Platform limits: live viewer evidence remains Windows 0.3.0a1. Linux/macOS viewer control is experimental and needs manual file selection. ARM64 package tests do not certify Raspberry Pi hardware. No remote viewer service is included.
Installation · Tools · Contribute
The wheel is for dedicated Python environments. SHA256SUMS.txt covers the supplied packages. Upgrades do not switch existing agent/viewer installations automatically.
Firestorm MCP 0.3.0a2 — portable setup alpha
Historical portable-setup alpha. For new installations, use 0.3.0a3.
This release added a common Python installer, POSIX launch scripts, generated JSON/Codex/VS Code configuration and offline diagnostics. Launch layouts cover Windows executables and experimental Linux wrappers/macOS app bundles. Native file selection remains Windows-only.
All 85 tests and eight CI jobs passed, including fresh source/wheel installs, MCP Inspector and dependency checks. An offline consumer review corrected four setup issues. Validation run.
Live viewer evidence remains the Windows 0.3.0a1 workflow. Linux/macOS desktop control and Raspberry Pi hardware were not tested. No remote viewer service is included.
These original assets are retained for reproducibility and rollback. SHA256SUMS.txt covers the source ZIP, wheel and source distribution; Python and Firestorm are separate prerequisites.
Firestorm mesh-preview skill 0.1.1
Optional instructions for agents using Firestorm MCP: mesh imports, LOD/physics controls, image comparisons, quote readback and cleanup.
Download firestorm-mesh-preview-0.1.1.zip and extract the complete folder. Add it through your host's skill mechanism, or ask a file-capable agent to read SKILL.md and its linked reference. Setup guide.
This update shortens the instructions and clarifies manual file selection on unsupported platforms. It requires a separately configured Firestorm MCP and does not change its permissions or restart the viewer.
Skill-format validation, four schema examples, deterministic packaging tests and an offline consumer review passed. The procedures retain Windows MCP 0.3.0a1 live evidence; no separate skill performance benchmark or universal host activation is claimed.
The ZIP contains the entrypoint, importer reference and MIT licence. A checksum file is supplied. People and AI agents are welcome to contribute clearer procedures and reproducible findings.
Firestorm mesh-preview skill 0.1.0
Historical mesh-preview skill. Use skill 0.1.1 for the current instructions.
The initial skill teaches scoped discovery, explicit LOD/physics files, source commits, checkbox keys, visual comparison, quote readback and cleanup with Firestorm MCP 0.3.0a1.
Format validation, four schema examples and 67 local tests passed, including deterministic packaging and private-content rejection. Two offline consumer reviews improved ownership and restoration guidance. Viewer procedures inherit the recorded Windows 7.2.4.80712 evidence; no separate skill speed benchmark was performed.
The original ZIP and checksum remain available. The ZIP contains SKILL.md, the importer reference and MIT licence. Install the MCP separately; adding instructions does not restart or reconfigure it.
Firestorm MCP 0.3.0a1 — Windows compatibility baseline
Historical Windows compatibility baseline. For new installations, use 0.3.0a3.
This release added MCP SDK v2, compact discovery with 43 workflow tools, structured results, strict arguments, catalog notifications and cancellation protection for queued calls. Viewer improvements include scoped UI search, unique button callbacks, targeted keys and bounded importer-camera input.
Breaking input change: ui_press_key requires a visible, enabled path. Read back selection and dependent controls; a selected label may need a separate commit.
Validation passed: 64 tests, four Windows Python 3.11–3.14 CI jobs, five wire-protocol revisions, legacy/current SDK clients, fresh installs and Inspector 2.6.0 with zero strict schema findings.
Live checks on Firestorm 7.2.4.80712 covered five LOD/physics inputs, selection readback, physics Analyze, preview zoom/orbit, captures, quote invalidation and cleanup. All five source hashes were preserved; no upload was submitted. The final installed helper was checked in a signed-in viewer with 18 APIs / 94 operations. Evidence and limits.
The original source ZIP, wheel, source distribution and checksums are retained for reproducibility and rollback. This baseline does not certify every host, viewer operation or simulator result.
Firestorm MCP 0.2.0a1 - Windows collaboration alpha
Historical Windows collaboration alpha. For new installations, use 0.3.0a3.
The initial package supplied 42 workflow tools, reference schemas for 94 discovered viewer operations, Windows setup/launch scripts, an MIT licence and contributor guides.
Validation passed: 35 local tests, Windows Python 3.11–3.13 CI, fresh wheel startup with 42 tools, package metadata and a 48-file source ZIP allowlist. The refactored launcher and expanded importer fields had not yet received live acceptance at this version.
The original source ZIP, wheel, source distribution and checksums are retained for reproducibility and rollback. Python and Firestorm are separate prerequisites. No viewer, user captures, session data or consumer assets are bundled.