fmind

Read Médéric Hurier's (Fmind) portfolio from the terminal.

uvx fmind whoami

The commands group sections of the website's profile API. Their names stay stable when the website's navigation or visual design changes.

--limit bounds articles and search results, and each project list (repositories and video series); --json prints a command's data instead of prose; --raw prints an article's Markdown source. Search also accepts --tag. Use --json and jq for other filtering.

--json works before or after the command: both fmind --json experiences and fmind experiences --json print the raw website section. Global --color / --no-color and --version go before the command. Colour is detected automatically, including NO_COLOR and redirected output. Use fmind --help or fmind <command> --help for options.

Output uses your terminal's default text colour and blue accent. Community roles and credentials use stacked entries with their published links, so they remain readable in narrow terminals. Links are never folded, so they stay copyable, and terminals that support OSC 8 hyperlinks make them clickable. Redirected output carries no trailing whitespace.

fmind search matches titles, summaries, tags and slugs; terms are ANDed, so each extra word narrows the result. --tag restricts to one of the site's own tags, case-insensitively; an unknown tag fails and lists the published ones.

fmind search "agent security"
fmind search --tag MLOps --limit 3

fmind read takes a slug, or enough of one to be unambiguous, and prints the article. In an interactive terminal it pages through $PAGER (default less, with LESS=FRX unless you set LESS), so short articles print directly; PAGER=cat disables paging. If the selected pager cannot start or exits unsuccessfully, the command reports an error; retry with PAGER=cat. Use --raw for the Markdown source.

slug="$(fmind articles --limit 1 --json | jq -r '.[0].slug')"
fmind read "$slug"
fmind read "$slug" --raw | glow -

--json turns every command into a data source, so the portfolio composes with the rest of the shell. JSON goes to stdout without colour or prose; errors go to stderr. With read --raw --json, JSON takes precedence and includes the Markdown in its markdown field.

fmind articles --limit 20 --json | jq -r '.[] | "\(.date[:10])  \(.title)"'
fmind search agent --json | jq -r '.[].url'
fmind certifications --json | jq -r '.certifications[] | select(.active) | .title'

whoami --json includes metadata, experience (a list containing the featured engagement, or an empty list), and services. certifications --json includes certifications, thesis, and specializations. projects --json includes open_source and youtube_series, each keeping the website's fields and bounded by --limit.

Migration to 2.0.0: projects --json now returns an object instead of one mixed array; change filters from .[] to .open_source[] or .youtube_series[]. --limit now bounds each list, so video series are no longer crowded out by repositories.

Migration to 1.0.0: certifications --json now returns an object instead of an array; change credential filters from .[] to .certifications[]. The experience field in whoami --json is additive. Update existing scripts before upgrading from 0.2.0.

uvx fmind whoami       # zero install
uv tool install fmind  # persistent
pipx install fmind     # alternative

Point it at another origin with FMIND_PROFILE_URL — useful against a local mise run watch server of the website:

FMIND_PROFILE_URL=http://127.0.0.1:8080/api/profile fmind whoami

Article reads and the MCP bridge follow the same origin, so one variable moves the whole CLI.

Install the optional MCP dependencies and let your agent host launch the bridge:

uv tool install 'fmind[mcp]'
fmind mcp

For a host that accepts JSON configuration:

{
  "mcpServers": {
    "fmind": {
      "command": "uvx",
      "args": ["--from", "fmind[mcp]", "fmind", "mcp"]
    }
  }
}

fmind mcp connects to https://www.fmind.dev/mcp using Streamable HTTP and forwards the website's tools, resources, and prompts over stdio. The website owns their names, schemas, and content. Set FMIND_PROFILE_URL in the host's environment to use /mcp on another website origin. No local data or response cache is used. The website answers MCP POST requests with JSON; GET/HEAD return 405 with Allow: POST because notification streams are not offered. Use /health for uptime probes. Requests have a 15-second deadline and remote HTTP responses are limited to 8 MiB.

Stdout carries only MCP messages; startup errors go to stderr. Portfolio output flags such as --json do not apply to mcp. The bridge forwards read-only request/response operations; subscriptions and server-initiated sampling or elicitation are not advertised.

From this checkout, run uv run --extra mcp fmind mcp. The optional extra uses the official MCP Python SDK; ordinary portfolio commands still need only Typer and Rich.

mise install       # install the pinned toolchain
mise run install   # sync the locked environment and install hooks
mise run format    # ruff imports and format, dprint
mise run check     # ruff, ty, actionlint, zizmor, gitleaks, dependency audit
mise run test      # offline pytest with branch coverage
mise run smoke     # exercise every command against the live site
mise run all       # format, check, test, build — the gate CI runs

The pre-commit hook runs checks without changing or staging files; run mise run format before committing. The pre-push hook runs the offline tests. CI runs the full gate on the pinned Python, tests Python 3.11–3.13 for compatibility, and tests the declared dependency floors with --resolution lowest-direct on Python 3.11.

A release tag must match the version in pyproject.toml (for example, v0.1.0). CD runs the same full gate for that tag and checks the live commands before publishing through PyPI Trusted Publishing. The live smoke check is a release check, not a pull-request gate.

MIT — © 2026 Médéric Hurier (Fmind).