Open Source on Brandon's Blog

mcp-snapotter

Sun, 30 Aug 2026 00:00:00 +0000

mcp-snapotter is an MCP server for SnapOtter, a self-hosted file-processing platform. It exposes SnapOtter’s 243 tools to any MCP-compatible client (Claude Code, Claude Desktop, etc.) through a clean, discoverable interface.

Instead of flooding the model’s context with 243 near-identical tools, it exposes eight generic ones and indexes the OpenAPI spec for discovery:

Tool	Purpose
`snapotter_list_tools`	Browse/search the catalog by section or keyword
`snapotter_describe_tool`	Show a tool’s documented settings fields
`snapotter_run_tool`	Run one tool on a file; awaits async jobs
`snapotter_run_pipeline`	Chain steps server-side in one pass
`snapotter_batch`	One tool over many files
`snapotter_upload_file`	Add a local file to the library
`snapotter_list_files`	List the SnapOtter library
`snapotter_get_file`	One file’s metadata and version history
`snapotter_download`	Fetch a downloadUrl to disk
`snapotter_health`	Connectivity check

Settings are discovered at call time from SnapOtter’s validation errors, so the server doesn’t need to model 243 settings schemas. Async tools (54 of them) are handled transparently with SSE-based progress polling.

Written in Python with strict mypy and ruff, MIT licensed.

Written with Claude / oh-my-pi running open models.

Tessera

Sat, 04 Jul 2026 00:00:00 +0000

Tessera answers: given where I live and the institutions I want access to, what is the cheapest legal combination of museum memberships that covers them?

Models the U.S. reciprocal-network landscape (NARM, ASTC, AHS, ROAM, MARP, ACM, AZA, Time Travelers), encodes per-network/per-institution exclusion rules as Datalog (Ascent), and solves membership-selection as weighted set-cover.

All-Rust stack, only generated JavaScript:

Leptos WASM SPA for the web UI
Ascent Datalog for rule encoding
Pure-Rust optimizer, zero C/C++ FFI
Local-first: user’s address never leaves the browser

Data is curated JSON with per-row provenance, shipped as rkyv zero-copy archives. Dual-licensed MIT / Apache 2.0.

Written with Claude / oh-my-pi running open models.

namor

Wed, 28 Jan 2026 00:00:00 +0000

Namor is a pronounceable-name generator that uses corpus-based phonotactics rather than random letters. Instead of producing gibberish like “xqztl,” it learns which sound combinations are plausible by analyzing real name patterns.

How it works:

Syllabification — breaks training names into onset, nucleus, and coda
Frequency extraction — counts how often each pattern appears in real names
Weighted sampling — generates new names by probabilistically combining patterns

Trained on 43 language corpora plus 26 mythological name sets. Ships as a Vite SPA with a standalone HTML build (~11KB, zero dependencies) that pre-computes frequency tables at build time. All computation happens in the browser — no server, no API calls.

TypeScript, Vite, corpus data from MatthiasWinkelmann/firstname-database. MIT licensed.

Written with Claude / oh-my-pi running open models.

asdf-teleport-ent

Tue, 15 Mar 2022 00:00:00 +0000

asdf-teleport-ent is an asdf and mise plugin for managing Teleport Enterprise CLI versions (tsh, tctl, teleport).

Built because I kept jumping between Teleport versions when testing upgrades and didn’t want to manage symlinks by hand. Works with both asdf and mise:

# asdf
asdf plugin add teleport-ent
asdf install teleport-ent 14.2.2
asdf set teleport-ent 14.2.2

# mise
mise plugin install teleport-ent
mise install teleport-ent@14.2.2
mise global teleport-ent@14.2.2

Plain shell — depends on nothing but bash, curl, and tar. Apache 2.0 licensed.

wordle-goggles

Thu, 27 Jan 2022 00:00:00 +0000

Wordle Goggles is a Wordle analyzer and solver that uses NLTK corpus data to filter out uncommon words from candidate lists. When Wordle was at peak hype in early 2022, this tool provided data-driven guesses instead of relying on human word-frequency intuition.

Uses the Reuters corpus via NLTK to score words by real-world usage frequency, narrowing the solution space faster than random guessing. Written as a single-file Python script — run it, feed in your Wordle results, and get the next best guess.

Python, NLTK. MIT licensed.

pathspec-ruby

Thu, 10 Jul 2014 00:00:00 +0000

pathspec-ruby is a Ruby gem that matches .gitignore-style path specifications. It follows the full gitignore syntax — negation patterns, directory-only rules, glob wildcards, and precedence ordering.

Ported from the Python pathspec library, it ships a CLI (pathspec-rb) and a Ruby API:

gitignore = PathSpec.from_filename('.gitignore')
gitignore.match 'coverage/index.html' # => true
gitignore.match 'important.txt' # => false (negated)
gitignore.match_tree '/src/repo' # returns matching file paths

Published as the pathspec gem on RubyGems. Supports Ruby 3.2 through 4.0. Apache 2.0 licensed.