Docs

How MESS works.

The short version: sign in with GitHub, wire your coding agent in once, say “sort out my mess.” Everything below is detail.

Start

Signing in is one GitHub click — no card, no form. You land on an empty ledger with the two ways to fill it.

Wire your agent is the primary path: your coding agent excavates this machine and keeps the ledger current forever after. Brain dump is the fallback for accounts no agent can see — you ramble, MESS parses.

Incomplete is fine. The ledger grows as you work, and it tracks its own gaps as open questions.

Wire your agent

Mint a key in Settings → Agent keys (shown once — it lands in the setup command automatically), then paste one command. One key per machine or agent is plenty.

$ claude mcp add --transport http --scope user mess https://mess.fyi/api/mcp \
    --header "Authorization: Bearer mess_sk_YOUR_KEY"

That covers Claude Code. Cursor, Claude.ai custom connectors, other MCP clients, and the plain REST surface are on the MCP server page. Once connected, the server itself tells your agent how to keep the ledger — logging new accounts unprompted, searching before creating.

The dig

The backfill prompt — “sort out my mess”, copyable from Settings → Agent keys — sends your agent through env files, CLI auth state, and configs on this machine. Every account it finds becomes a labeled row.

It works in passes — first sweeps routinely surface only half of what's really there. Unknown fields are logged anyway and become open questions. Secret values never move and never touch MESS.

If the dig hits your plan's cap, nothing is lost: the agent keeps its findings and reports what didn't fit instead of dropping it.

Brain dump

For everything an agent can't see from a terminal — the client's registrar, the Stripe you opened in 2019. Ramble in plain language; MESS parses it into rows you review before anything saves.

> there's the main supabase on my gmail, another one for acme that
  billing owns, stripe for acme lives in 1password, and I think
  there's still a neon db from that hackathon somewhere

⎿ 4 accounts parsed — review before saving

Open questions

Derived from your rows — missing owner email, credentials location, cost, or dashboard URL. No AI involved and nothing to configure: fix the row and the question disappears.

Each question comes with a copyable prompt. Hand it to your agent and the answer lands back in the ledger — the agent that deploys with an account can usually read who owns it.

Handoff

One click renders the ledger to a markdown document — deterministic on purpose, no AI in the pipeline, because a handoff has to be trustworthy. Scope it to one client or one project.

Provider, label, owning email, plan, where the credentials live — grouped by role, with cancelled rows noted. It's also your no-lock-in guarantee: the whole ledger walks out the door as plain text whenever you want.

Keys & limits

Agent keys are bearer tokens (mess_sk_…), shown once at mint, stored hashed, revocable any time in Settings → Agent keys.

Each key gets 240 tool calls per hour on the MCP surface and the same again on REST. A rate-limited agent is told when to retry — nothing already logged is lost.

MESS stores which accounts exist and where their credentials live — never the credentials themselves. There is no field for a secret value. The map, not the vault.

Plans

Free is the full product to 20 accounts — agents, search, handoff, everything. $40/year removes the cap. That is the entire price list.

Billing lives in Settings → Billing (Stripe Customer Portal — cancel any time, the ledger stays readable either way). Details on the pricing page.

Troubleshooting

The short list — if yours isn't on it, hello@mess.fyi reaches a human.

Sign-in bounced back to login

OAuth round-trips must finish in the browser they started in. Start again from mess.fyi/login in one browser and it goes through.

Agent reports rate limited

The 240/hour per-key budget reset time is in the error message. Everything already logged is safe; the agent is told to pause, not abandon findings.

Hit the account cap mid-dig

Nothing breaks and nothing is dropped — the agent reports what didn't fit. Upgrading (mess.fyi/upgrade) takes the cap off and the next dig logs the rest.

Lost an agent key

Keys are shown once and stored hashed, so there is no reveal. Revoke it in Settings → Agent keys and mint a new one — the setup command updates itself.

Duplicate rows after a dig

Merge by hand or tell your agent to consolidate: update the row you keep, mark the duplicate cancelled. There is no delete — cancelled rows drop out of search but keep their history.