fons

<picture> <source media="(prefers-color-scheme: dark)" srcset="assets/banner-dark.png"> <source media="(prefers-color-scheme: light)" srcset="assets/banner-light.png"> <img alt="Fons — the living profile that gets your company investment ready" src="assets/banner-light.png"> </picture> <p align="center"> <b>Fons gives your company a living agentic profile that gets it ready for investment.</b><br> One profile that works as your data room, your due-diligence evidence, and the way investors and their agents discover you. </p> <p align="center"> <a href="https://fons.vc"><img src="https://img.shields.io/badge/fons.vc-live-0a0a0a?style=flat-square" alt="fons.vc"></a> <a href="#install"><img src="https://img.shields.io/badge/Claude_Code-plugin-0a0a0a?style=flat-square" alt="Claude Code plugin"></a> <a href="https://modelcontextprotocol.io"><img src="https://img.shields.io/badge/MCP-streamable_http-0a0a0a?style=flat-square" alt="MCP"></a> <a href="LICENSE"><img src="https://img.shields.io/badge/licence-Apache--2.0-0a0a0a?style=flat-square" alt="Apache-2.0"></a> <a href="https://discord.gg/HVXjkCxgcj"><img src="https://img.shields.io/badge/Discord-join-0a0a0a?style=flat-square&logo=discord&logoColor=white" alt="Discord"></a> <a href="https://x.com/fons_vc"><img src="https://img.shields.io/badge/X-@fons__vc-0a0a0a?style=flat-square&logo=x&logoColor=white" alt="X"></a> </p> --- This repository is the **agent surface of Fons**: the Fons Skill, the MCP wiring it depends on, and the `fons` CLI. Install it and your AI assistant can read and update your Fons profile, your company, your product, your key metrics and your data room — from wherever you already work, without a browser tab and without a single credential passing through the chat. The Fons platform itself (the web app, the scoring engine, the database) is **not** open source and does not live here. What lives here is everything you need to *talk to it*. ## Contents - [What your agent can do](#what-your-agent-can-do) - [Install](#install) - [How it authenticates](#how-it-authenticates) - [The CLI](#the-cli) - [Readable without an account](#readable-without-an-account) - [What it deliberately cannot do](#what-it-deliberately-cannot-do) - [Architecture](#architecture) - [Contributing](#contributing) - [Follow us](#follow-us) - [Security](#security) - [Licence](#licence) ## What your agent can do Twenty tools, all scoped to **your own** account. Authorization is decided by the Fons server on every single call, so the agent can never exceed the permissions you already have. | | Tools | |---|---| | **Identity** | `whoami` · `get_profile` · `update_profile` | | **Company** | `get_company` · `update_company` | | **Product** | `get_product` · `update_product` — publishes your public `/p/` page | | **Evidence** | `get_metrics` · `update_metric` · `delete_metric` · `list_documents` | | **Readiness** | `get_company_readiness` — what's captured, unclear, missing | | **Audit trail** | `get_feed` · `post_feed_note` — append-only, investor-facing | | **Data sources** | `list_company_connections` · `sync_company_connection` · `connect_company_source` | | **Verified accounts** | `list_connections` · `connect_account` · `disconnect_account` | The Skill is what makes those tools *useful*. It teaches the agent the field vocabulary, the validators, which edits are permanent, which ones void a verification badge, and — when something can only be done in a browser — the exact URL that opens the right control rather than a shrug towards the homepage. ``` you › get my company investment ready on fons → reads your readiness checklist, drafts the gaps with you, records the metrics, tells you which evidence is still missing, and hands you a link to connect Stripe so your revenue numbers carry verified provenance instead of your word for it. ``` ## Install You need a Fons account — free at [fons.vc/app](https://fons.vc/app). A personal profile costs nothing. A company profile is Fons Pro (£49/month per company). ### Claude Code — one step ```bash /plugin marketplace add fons-vc/fons /plugin install fons@fons ``` That installs the Skill **and** wires up the MCP server. The first tool call opens a browser consent screen; approve it and you're connected. ### Every other agent One URL, everywhere: ``` https://fons.vc/mcp ``` It's a remote **Streamable HTTP** MCP server with **OAuth 2.1, PKCE and dynamic client registration**, so in most clients you paste that URL and the browser does the rest — no key, no token, no callback URL to configure. <details open> <summary><b>Claude</b> — claude.ai, Claude Desktop, Claude Code</summary> - **claude.ai / Desktop:** Settings → Connectors → **Add custom connector** → paste the URL. (Custom connectors need a paid plan.) - **Claude Code, without the plugin:** `claude mcp add --transport http fons https://fons.vc/mcp` </details> <details> <summary><b>ChatGPT</b> — Developer mode</summary> Paid plans (Plus, Pro, Business, Enterprise, Edu), **on the web**. Not available on Free. On Business/Enterprise your workspace admin has to enable developer mode first. 1. **Settings → Security and login → Developer mode** — turn it on. 2. Go to **Settings → Plugins** (or `chatgpt.com/plugins`) → **+**. 3. Name it `Fons`, and set the **MCP server URL** to `https://fons.vc/mcp`. **Create**. 4. In a chat: **＋ → Developer mode → Fons**. You enable it per conversation. ChatGPT asks you to confirm each write the first time — that's its behaviour, not ours, and we think it's the right one. (OpenAI moves these menu labels around; if yours differ, look for "developer mode".) </details> <details> <summary><b>OpenAI Codex</b></summary> Add to `~/.codex/config.toml`: ```toml [mcp_servers.fons] url = "https://fons.vc/mcp" auth = "oauth" ``` Then `codex mcp login fons`. </details> <details> <summary><b>Cursor</b></summary> [](https://cursor.com/en/install-mcp?name=fons&config=eyJ1cmwiOiJodHRwczovL2ZvbnMudmMvbWNwIn0%3D) Or by hand, in `~/.cursor/mcp.json`: ```json { "mcpServers": { "fons": { "url": "https://fons.vc/mcp" } } } ``` </details> <details> <summary><b>VS Code</b> — GitHub Copilot agent mode</summary> Command Palette → **MCP: Add Server**, or write `.vscode/mcp.json` (note the key is `servers`): ```json { "servers": { "fons": { "type": "http", "url": "https://fons.vc/mcp" } } } ``` Works on every Copilot plan including Free. Copilot Business/Enterprise members need their org to allow MCP servers. </details> <details> <summary><b>Gemini CLI</b></summary> ```bash gemini mcp add --transport http fons https://fons.vc/mcp -s user ``` Then `/mcp auth fons` inside the CLI. It needs a local browser for the consent step, so it won't complete over a bare SSH session. **The Gemini app (gemini.google.com) can't do this** — its connectors are a fixed Google-curated list with no way to add your own. Use the CLI, or just [read the `.md`](#readable-without-an-account). </details> <details> <summary><b>Grok</b></summary> **grok.com/connectors → New Connector → Custom** → paste `https://fons.vc/mcp` → complete the sign-in it prompts for. On Grok Business/Enterprise a team admin adds the connector first. </details> <details> <summary><b>Zed</b></summary> Settings → AI → MCP Servers → **Add Remote Server**, or in `settings.json`: ```json { "context_servers": { "fons": { "url": "https://fons.vc/mcp" } } } ``` Leave the headers out and Zed runs the OAuth flow for you. </details> <details> <summary><b>Windsurf</b> (now Devin Desktop)</summary> `~/.codeium/windsurf/mcp_config.json` — the remote key here is `serverUrl`, not `url`: ```json { "mcpServers": { "fons": { "serverUrl": "https://fons.vc/mcp" } } } ``` </details> <details> <summary><b>Llama, Mistral, DeepSeek, Qwen — or anything you run yourself</b></summary> These are **models, not agent hosts** — they have no connector settings of their own. Run them inside an MCP client and point that client at Fons. Two that handle our browser sign-in properly: - **LibreChat** — per-user OAuth, and users can add the server from the UI. In `librechat.yaml`: `type: streamable-http`, `url: https://fons.vc/mcp`. - **Open WebUI** (v0.6.31+) — Admin Settings → External Tools → **+** → MCP (Streamable HTTP) → URL → Auth: **OAuth 2.1 (DCR)**. Admin-only, and you must enable it per chat for the sign-in to fire. **Ollama** runs the model, not the tools — it has no MCP client, so you can't point it at Fons directly. Use it as the model *behind* LibreChat or Open WebUI. Likewise **OpenRouter**: it's an inference router, not an agent host. Neither is a gap in Fons; it's just not what those tools do. </details> <details> <summary><b>Not working yet — and we'd rather say so</b></summary> | | Why | |---|---| | **Perplexity** | It *has* custom remote connectors (Pro/Max), but its OAuth registration currently insists on a `client_secret`. Fons registers clients as **public** ones — no secret, PKCE instead — which is the correct and safer design for an app that can't keep a secret. Perplexity rejects that today ([open bug](https://community.perplexity.ai/t/custom-mcp-connector-fails-with-did-not-return-a-client-secret-for-rfc-7591-compliant-public-client-registrations/5172)). We tested it; it fails. We'll list it the day it works. | | **Microsoft 365 Copilot** | Custom MCP connectors are admin-configured, need a pre-registered client secret, and are documented for **read-only** tools. **Copilot Studio** does support our sign-in (Tools → New tool → MCP → OAuth 2.0 → Dynamic discovery) if you're building an agent there. The **consumer** Copilot app has no custom connectors at all. | | **Any model, no setup** | All of these can still *read* a Fons profile — see below. | </details> The Skill is what turns those tools into good behaviour. Outside Claude Code, copy [`plugins/fons/skills/fons`](plugins/fons/skills/fons) into your agent's skills directory (`~/.claude/skills/fons` for Claude Code), or just paste `SKILL.md` into the conversation — it's plain Markdown and it works as context anywhere. ## How it authenticates **No credential ever passes through the chat.** Not an API key, not a token, not a password. This is the whole point, so it is worth being precise about: 1. Your agent calls a Fons tool. The server replies `401` with an OAuth challenge. 2. Your agent opens a **browser consent screen inside your logged-in Fons session** — you approve, as yourself, in a place an agent cannot reach. 3. The agent receives a scoped token and acts **as you**. Every call is authorized server-side against the same rules the website uses. Row-level security is the boundary, not the client. The same pattern covers anything that requires proof of ownership. Verifying a LinkedIn account or connecting Stripe returns a **bridge URL** you open in your own browser — the agent hands you the door; it never holds the key. If you paste a secret at it anyway, the Skill is written to refuse it and tell you to rotate it. ## The CLI ```bash npm install -g fons-cli # or: curl -fsSL https://fons.vc/install.sh | sh (Node 18+) fons login # OAuth, same browser consent, no tokens to copy fons profile get fons profile set --headline "Founder, Acme — B2B infrastructure" fons company # your company + data room fons readiness # the evidence checklist fons product set --tagline "…" fons metrics fons connectors ``` The token is stored mode-`0600` at `~/.config/fons`. Same auth, same server-side authorization, same refusals. See [`cli/`](cli). ## Readable without an account Every public Fons profile ships an **agent-readable Markdown twin** at the same address — with YAML front-matter and JSON-LD, so a model gets structured facts instead of a scraped page. No auth, no SDK, no API key. Try it right now: ```bash curl https://fons.vc/c/fons-vc.md # a company (this one is ours) curl https://fons.vc/p/<slug>.md # a product curl https://fons.vc/<handle>.md # a founder ``` That's the discovery half of Fons, and it works with **every** model — no connector, no setup. Paste the URL into ChatGPT, Gemini, Grok, Perplexity, Copilot, or a local Llama, and ask about the company; anything that can read a web page can read a Fons profile. It is also why the profiles are worth keeping accurate: an investor's agent doing background reading will find the `.md`, not your pitch deck. The MCP tools in this repo are for **your own** data. The `.md` twins are how any agent reads **everybody else's**. ## What it deliberately cannot do An honest surface is more useful than an omnipotent one. The agent **cannot**: - **Assert a verification it didn't earn.** Verified connections, trust badges and `connected` metric provenance are written only by a live connector or a proven browser session. An agent can type a number into your data room; it cannot make it *verified*. That distinction is the product. - **See or invent a score.** Fons reports readiness as statuses and bands — never a number. The scoring internals are not exposed to any agent, ever. - **Read your documents.** `list_documents` returns names and dates. File contents never flow through a third-party AI session. - **Forge your history.** The company feed is append-only and verification events are server-written, so an audit trail cannot be manufactured from a chat window. - **Exceed your permissions.** A view-only teammate gets an honest `403`, not a workaround. ## Architecture ``` your agent this repo fons.vc ┌───────────────┐ ┌────────────────────┐ ┌──────────────────┐ │ Claude Code │ skill ───▶ │ SKILL.md │ │ /mcp │ │ claude.ai │ │ references/ │ │ OAuth 2.1 + PKCE│ │ Cursor, … │ tools ───▶ │ .mcp.json ─────────┼───────▶│ ↓ │ └───────────────┘ │ cli/ │ │ RLS authorizes │ └────────────────────┘ │ every call │ ▲ └────────┬─────────┘ │ │ └────────── browser consent (you, logged in) ◀────────────────┘ ``` The Skill is instructions; the MCP server is the door; **the database is the boundary**. A fork of this repo still authenticates as you, through your consent, under your permissions — which is why we can publish it. ## Contributing Issues and pull requests are welcome, particularly on the Skill's guidance: if it gave your agent a bad instruction, that's a bug worth reporting. See [CONTRIBUTING.md](CONTRIBUTING.md). Note that `SKILL.md`, `references/` and `cli/` are generated from the Fons monorepo — we accept patches here and land them upstream. ## Follow us - **Discord** — [discord.gg/HVXjkCxgcj](https://discord.gg/HVXjkCxgcj). Where we talk about what agents should and shouldn't be trusted to do with a company's own data. Bring the transcript where your agent got it wrong; that's the useful conversation. - **X** — [@fons_vc](https://x.com/fons_vc). - **Web** — [fons.vc](https://fons.vc). Every public profile there has a `.md` twin, including [ours](https://fons.vc/c/fons-vc.md). ## Security Found a vulnerability? **Do not open a public issue.** See [SECURITY.md](SECURITY.md) — report it to [admin@fons.vc](mailto:admin@fons.vc). ## Licence [Apache-2.0](LICENSE). Fons, the Fons logo and fons.vc are trademarks of Fons — the licence grants no rights to them (§6), so a fork must not present itself as Fons.