MCP Server
Connect Bounce Watch to your assistant
Ask which companies just raised, who started hiring, what changed on a homepage last Tuesday — and get answers with the date we saw each line. There are two ways in, and the one that suits you depends on whether you write code.
Which Dutch companies raised in the last 60 days and are hiring sales?
search_signals
Four companies match. The strongest is Ventory:
- Closed a €2.65M round on 12 May
- Posted 6 roles, 3 of them commercial, from 28 May
- Announced expansion plans 1 May
The raise and the sales hiring are two weeks apart — that ordering usually means the go-to-market spend is being committed now.
Setup
Two values and you're done.
A server URL and your API key. Claude Code takes them in one command; Cursor and VS Code install in one click from your panel; Claude Desktop through a copy-paste config. Everything else speaks plain JSON-RPC over a single POST.
This is the developer path onto the API server. If what you want is your own tracked companies inside Claude or ChatGPT, that server takes no key at all — you sign in and approve a screen.
- 01
Create an account and grab your API key — trial credits included.
- 02
Open
API Panel → AI Agentsand copy the config for your client, or hit the one-click install. - 03
Ask your agent a question about timing.
{ "mcpServers": { "bouncewatch": { "url": "https://bouncewatch.com/api/v1/mcp", "headers": { "Authorization": "Bearer YOUR_API_KEY" } } } }
MCP server
Or ask it yourself, from Claude or ChatGPT.
Bounce Watch runs as an MCP server. Connect it once and Claude, ChatGPT, Cursor or your own agent can ask the record anything, in plain language. Three steps, and you never touch a key or a config file.
-
01
Open Settings → Connectors → Add custom connector.
Claude asks for a paid plan at this step — theirs, not ours.
-
02
Paste this link.
https://api.bouncewatch.com/api/v1/mcp -
03
Sign in and approve. That is it — nothing to paste back.
Bounce Watch
wants to connect to your assistant
- Read company signalsFunding, hiring, partnerships, expansion and the rest of the record
- Request a fresh lookCheck a company again when we have not looked recently
- Spend credits from this accountThe same pool as your API key
We never see what you ask your assistant — only the questions it sends here.
On Claude's free plan, — same three steps, nothing to install.
One switch is called developer mode, and nothing technical follows it. You will not write any code, it changes nothing else in ChatGPT, and you can turn it off in the same place.
-
01
Settings → Apps → Advanced settings → turn on developer mode.
-
02
Add a connector and paste this link.
https://api.bouncewatch.com/api/v1/mcp -
03
Sign in and approve.
Bounce Watch
wants to connect to your assistant
- Read company signalsFunding, hiring, partnerships, expansion and the rest of the record
- Request a fresh lookCheck a company again when we have not looked recently
- Spend credits from this accountThe same pool as your API key
We never see what you ask your assistant — only the questions it sends here.
Add this to your MCP config and sign in when it asks.
{ "mcpServers": { "bouncewatch": { "type": "http", "url": "https://api.bouncewatch.com/api/v1/mcp" } } }
One command.
claude mcp add --transport http bouncewatch \
https://api.bouncewatch.com/api/v1/mcp
It opens a browser to sign in the first time. Add -s user to make it available in every project.
Any client that speaks MCP over HTTP. The first call costs nothing and lists every tool your agent can use.
curl -s https://api.bouncewatch.com/api/v1/mcp \ -H "Authorization: Bearer YOUR_KEY" \ -H "Content-Type: application/json" \ -d '{"jsonrpc":"2.0","id":1,"method":"tools/list"}'
Prefer a key to a sign-in? Read how keys work — no card, and it works on every tab above.
Free to connect — 2,500 credits, no card.
What your agent can ask
15 tools. Two of them are the reason you are here.
Company lookup you can get anywhere. search_signals
answers a question no company database can — which companies did something recently,
and when. watch_company answers the
one that comes next: tell me when it happens again, without being asked.
| Tool | What it does | Credits |
|---|---|---|
| search_signals | The latest signals across the whole index: which companies did something recently, what it was, and when. | 12 |
| search_companies | Finds companies by firmographics — country, headcount, funding stage — within the set BounceWatch actively observes, most recently refreshed first. | 5 |
| find_company | Finds a company in the BounceWatch index by name, and returns its domain — which is what every other company tool needs. | 3 |
| get_company | Firmographic profile of one company from the BounceWatch index: identity, location, headcount, business model and funding history. | 10 |
| get_company_signals | Returns the dated signal timeline for one company: funding, hiring, partnerships, expansion, product launches, leadership changes and risk events. | 8 |
| refresh_company | Queues a fresh scan of one company and returns a batch id to poll with get_refresh_status. Works for companies already in the index and for domains we have never seen, which is how you add one. | 36 |
| get_refresh_status | Checks a scan queued by refresh_company, and WAITS for it. | Free |
| watch_company | Registers a standing watch on one company, so you find out what happened there without asking again. This is the only tool that keeps looking on its own between sessions. | 5 |
| check_watches | Returns what has happened at the companies THIS KEY ALREADY WATCHES, since the last time you asked. | Free |
| list_watchlists | Lists the watchlists on this account: named sets of companies, usually one per client or project, with how many companies each holds and what has been spent on it. | Free |
| manage_watchlist | Creates a watchlist, or renames, archives or restores one. | Free |
| update_watchlist_companies | Adds companies to a watchlist or removes them, by domain. | 62 per company |
| scan_watchlist | Scans the companies in a watchlist, or cancels a scan in progress. | 62 per company |
| get_watchlist | Returns one watchlist as a map of where to look: each company with how many signals it has in the window and the strongest of them, plus when it was last scanned and the progress of any scan in progress. It carries no signal bodies; read a company worth a closer look with get_company_signals. | Free |
| get_signal_taxonomy | Lists every signal type BounceWatch detects, grouped by category, with what each one means and whether it is an announcement or an unconfirmed inference. | Free |
The part nobody advertises
It tells your agent what it doesn't know.
The dangerous failure of a signal API is not a wrong answer. It is an empty one.
An agent reads signals: [],
concludes the company has gone quiet, and tells your team to deprioritise an account
that is in fact perfectly alive — we simply had not looked at it since March.
So every answer carries how recently we observed that company, and a flag saying whether the absence of signals means anything at all. When it doesn't, the agent is told to say so rather than guess.
"coverage": { "status": "stale", "observed_at": "2026-05-01", "age_days": 86, "signal_absence_is_meaningful": false, "interpretation": "Last observed 86 days ago. This is a coverage gap, not a signal history. Do not call this company quiet on the strength of this data — refresh first." }
Running it
What it costs, and who can open it.
Every call is metered against the same credit balance as the API, and both servers are authenticated — a key for code, a sign-in for everything else. Most MCP servers listed publicly are neither.
28 Jul 2026
6 Aug 2026
API key
OAuth
Metered credits
Plan-gated
Every tool call is metered
search_signals 12 · get_company 10 · refresh_company 36 · polling free
An agent that runs nightly bills nightly — on a product we already sell.
And ours can wake the agent
watch_company and
check_watches shipped in
August 2026. The rest of this market stops at read: your agent asks, repeatedly,
and pays for every question it did not need to ask.
Live run, 6 August: a company last scanned 693 days earlier → 11 signals and a scored card in under three minutes.
Two ways to authenticate, on purpose
A key is right for code: it goes in an environment variable and it never expires on you mid-run. It is wrong for everybody else. Claude Desktop and claude.ai will not take one, and a salesperson is never going to edit a JSON config to reach their own pipeline.
So the Signal Tracker server speaks OAuth: you paste a URL, sign in, approve a screen that names what the assistant will be able to see, and you are connected. Nothing to copy back.
See the one-click connection-
Tokens are bound to one resource. A Signal Tracker grant cannot open the API server, and the reverse cannot happen either.
-
41% of listed MCP servers have no authentication at all. Of the ones that do, 8.5% use OAuth. We are in that 8.5%, and wiring an open server into anything that touches your pipeline is a security decision rather than an integration one.
Sources · Zuplo survey of 17,000+ listed MCP servers, Feb 2026 · BlueRock analysis of ~7,000 public servers
Push, not pull
You don't search for a funding round. You wait for one.
The signals worth the most are the ones that happen least. Across every company we watch, a quarter holds a few hundred funding rounds and a handful of shutdowns — because that is how often those things happen, not how closely we are looking.
Which means searching for them is the wrong instrument. The odds one lands inside the window your agent happens to ask about are poor, and a quiet result reads like an empty index rather than a calm quarter.
So the agent registers the interest once and we do the waiting.
watch_company on the accounts that
matter; check_watches returns only
what is new since it last asked, and costs nothing. If your agent runs on a server, the
webhook reaches it the moment the signal lands.
check_watches() 3 new since Thursday mollie.com recently_funded Aug 8 bunq.com key_hire_announced Aug 9 framer.com expansion_announced Aug 10
Nothing was polled and nothing was re-read. The protocol only lets a server answer, so the registration is a normal tool call and the wake-up arrives another way — an inbox the agent drains, or a signed webhook.
Nothing to prompt-engineer
The jobs come built in.
Connecting the server also installs a set of workflows. Pick one from your client's menu instead of working out which tools to chain — each already knows the right sequence.
Same key, same credits, no extra subscription.
The MCP server is a second door onto the Data Enrichment API. If you already have a key, it works right now — nothing to buy, nothing to migrate.
