7Maps
7Maps checks every remote MCP server in the official registry and signs every check. Agents ask which server works and which tool fits, before connecting.
Hosted MCP Server
npx add-mcp 'https://7it.co.il/7maps/mcp?via=mcpservers'Installs into Claude Code, Codex, Cursor and more
Documentation
Signed history of which MCP servers actually work.
… answering now
servers with tools
20,000+ on the map
from the official registry
… days signed
since 2 October 2026
Reading the mapIndex: loadingDaily check: loading
Updated 4 Oct 2026: tools whose name and description only describe reading are now marked read-only when nothing suggests they change data (about one in ten tools). This is a change in how levels are estimated, not in the servers, so it does not show as a change in watch or the feeds. The changelog.
The signed record as a course on a chart: one point for each signed day, chained to the day before, the newest root marked and the next root open.
History that cannot be backfilled
A check made today can only be proven tomorrow. That is why the record starts early and never stops.
| … | servers with 30 or more days of signed history on the map now |
|---|---|
| … | servers with 90 or more days counted as history grows |
| … | servers with a full year or more counted as history grows |
| What is signed | Every check: the daily check of every server, the hourly check of the most used ones, the five-minute checks of servers whose owners verified them, and the checks of new servers under observation. The daily check of every server is in the record from 2 October 2026; the other checks joined on 5 October 2026. |
|---|---|
| How | Each UTC day becomes one root: a SHA-256 hash of every file of that day, chained to the root of the day before and signed with Ed25519. A day that has a root is never signed again. |
| Check it yourself | Pick any day and your browser recomputes the root and verifies the signature with the published key. |
Paid servers, priced and checked
Agents now pay servers per call. Before an agent pays, 7Maps tells it which servers are up, which have a record of answering, and what each one says it charges.
Counting started on 6 October 2026. The first count covers the map one part at a time.
| x402 | 0 |
|---|
Fig. 1 Servers that take agent payments, by protocolcounting
| 0 | servers that accept agent payments of the servers counted |
|---|---|
| 0 | payment protocols declared x402, MPP, ACP and others |
| 0% | of them publish a price per call shown to agents as declared |
| 0 | of them answering at the newest check a payment needs a server that answers |
Allowance at no cost, among servers that accept payments
Declare some calls at no cost 0 Declare no allowance 0
How an agent decides, in one call to 7Maps
| Up | The server answered at its newest check. |
|---|---|
| Reliable | A signed record of how often it answered. |
| Price | What the server declares it charges per call. |
| Choose | The agent picks, then pays the server directly. |
| For agents | find_tool and route take an optional payment filter, and every paid server in an answer carries its declared price next to its status. 7Maps sits before the payment: it never handles money and never takes a cut. |
|---|---|
| For server owners | When a verified server that takes agent payments stops showing its payment details while it still answers, the owner gets an alert in plain words. A broken payment request means every paid call is lost. |
| What is counted | Only what servers declare in the answers 7Maps already receives and in their registry entries. 7Maps never pays and never calls a tool. Unclear declarations count as unknown, never as a guess. |
The map in numbers
Every server at its newest check, read live while this page is open.
| Answers with tools | … | |
|---|---|---|
| Sign-in first | … | |
| Payment first | … | |
| Not answering | … |
Fig. 2 Servers by status at their newest checklive
| 10,000+ | answering with tools at their newest check loading |
|---|---|
| 20,000+ | servers checked in the last 24 hours daily check |
| 200,000+ | tools with an estimated risk level, at the last daily check read-only, changes data, high risk |
| 5,000+ | tokens an agent loads, on average, just to read one server's tool list what one 7Maps answer replaces |
Rounded figures until the live numbers load.
Where the servers on the map are hosted
Counts by country, computed once a day.
Fewer serversmore servers Ring: global edge networks (one address answered from many countries) Magenta point and courses: our daily checks go out from US East Country of each server's address: IP geolocation by DB-IP (CC BY 4.0). Outlines: Natural Earth.
World map shaded by the number of MCP servers whose address is registered in each country. The list beside the map has the same numbers.
Take the controls
Send an agent to find a tool, switch off parts of the map, or set what an agent may use on its own. Dispatch runs a real search over every tool on the map. The outage and policy consoles are simulations on the public counts.
Your dispatches 0
Try
Job for the agent
query postgres
any
One press, one real search. The pulse on the map goes out over the whole map: where each server is hosted is not published, so no result is placed on it.
Readout
Waiting for a dispatch
Pick a job and press Dispatch. Results come from the public tool search, with each server's real status and handshake time.
Changed today
From the daily checks and the hourly check of the most used servers. Every server on the live map.
- Loading.
Find a tool
Describe the job. Results come from every tool of every server on the map, with what the tool can do and whether its server answers now.
Levels are automated estimates from each tool's public description and can be wrong. How good is this search: the tool search evaluation.
Status board
The most used servers, and today's numbers over every server.
| … | most used servers in the hourly check |
|---|---|
| … | of them not open at that check |
| … | answering with tools (State of MCP uptime) |
| … | median handshake |
| … | median tool list, in tokens |
| … | of sign-in servers publish OAuth metadata |
Today's numbers over every server: State of MCP uptime. Monthly numbers, by category, with a dataset to cite: the monthly report. Ranked by data only: the most reliable servers by category. How 7Maps sits next to other services: compare.
7Maps news
What changed for MCP servers and the agents that use them, in brief and with sources.
What changed for MCP servers and the agents that use them, in brief and with sources
Put the map inside your agent
Streamable HTTP, no sign-in. Listed in the official MCP registry as io.github.XLSV777/7maps.
Claude Code plugin, with a skill that checks servers before connecting: /plugin marketplace add XLSV777/7maps then /plugin install 7maps@7maps. Gemini CLI: gemini extensions install https://github.com/XLSV777/7maps. The plugin page covers install, data and privacy, and uninstall.
Claude Code, one line
claude mcp add --transport http 7maps https://7it.co.il/7maps/mcp?via=claude-code
Claude and ChatGPT apps: add a custom connector with this address
https://7it.co.il/7maps/mcp
{
"mcpServers": {
"7maps": { "type": "http", "url": "https://7it.co.il/7maps/mcp" }
}
}
Fig. 3 Any MCP clientconfig
Optional: create a personal key (no account, no email) and send it as the X-7Maps-Key header to see what 7Maps did for you. The plugin asks for it; leaving it empty works the same.
Machine description: /.well-known/7maps.json.
Why agents ask the map first
To find out what a server can do, an agent has to connect and load its whole tool list. Across more than 20,000 servers that is a lot of tokens and time spent on servers that turn out to be down, behind a sign-in, or wrong for the job. 7Maps has already done that work. One question returns the server and the tool, with its risk level and what changed since a person approved it.
Every answer gives an estimate of the tokens it saved compared with finding the same thing out without 7Maps, and names its source. With an optional personal key, your own dashboard adds them up: calls, tokens saved and failures avoided.
Make it a habit for your agent
Installed tools are only used when the agent thinks of them.
Paste this into CLAUDE.md, AGENTS.md or .cursor/rules so it asks 7Maps every time:
## MCP servers
- Before connecting to an MCP server you have not used today, call 7Maps road_conditions
with its address. Do not connect if it is down, or if its tools changed since a person
approved it, until the person approves again.
- When you need a tool and do not know which MCP server has it, call 7Maps find_tool (or
route) with the task in plain words, and use the server and tool it returns.
- After calling a server 7Maps pointed you to, pass last_trip on your next 7Maps call.
Fig. 4 The rule
What an agent can ask
find_tool | For a job: search every tool on the map, with what each can do and whether its server answers now. No charge. |
|---|---|
route | For a task: the open server and tool that can do it with the least access, plus alternatives and servers to avoid right now. |
road_conditions | Before connecting: status, speed, every tool's risk level, the last change and the agents' rating. |
watch | Before reusing a server a person approved: did its tools change or get riskier since that date. |
tool_card | One tool's exact schema and risk, without loading the whole list. |
preflight | Will these arguments pass the tool's schema, and is the server up. |
verify_lock · changes_since | Does a server still match what was approved, and what changed across a set of servers. |
Use it every day without a wallet
find_tool, about_7maps and report_road are never charged.The first 20 calls a day of route and road_conditions are not charged.Owners pay nothing to put a server on the map, verify it, get alerts or read my_server_report.
Past the daily allowance, a call is paid in USDC on Base through x402, from the wallet the agent was given; list_paid_tools returns current prices. about_7maps shows a real example of the savings before anything is paid. Agents that report how a call went (report_road) build the ratings, and paying reporters earn calls back.
Own an MCP server?
What owners get at no charge, and what the owner plans add: for server owners.
| Not on the map yet | Small businesses and new servers can put their server on the map. It is observed for 7 days, then agents can be routed to it. Placement cannot be bought. |
|---|---|
| Already on it | Every server has a page at 7it.co.il/7maps/s/<host>/<path> with what was observed and when, a README badge and a live status card to embed (7it.co.il/7maps/widget/<host>/<path>.js). |
| Prefer not to be listed | How 7Maps-bot works and how to opt out. Removal is never charged. |
Sensors for gateways
If you run an MCP gateway, it already sees how real calls to many servers go.
An opt-in sensor sends 7Maps the outcome of each tool call your gateway forwards. Outcomes from many gateways become the ratings and the live incident flags that every agent reads before it connects.
| What is sent | The server address (no query string or credentials) or registry name, worked or failed, one of seven failure reasons, milliseconds, the tool name and the time. Plus a random sensor id, the gateway name and the sensor version. |
|---|---|
| What is never sent | Arguments, results, prompts, user ids, keys or tokens. 7Maps refuses a batch with any other field, keeps only servers on the public map, and does not store IP addresses. |
| Off until you switch it on | Nothing is sent until you set SEVENMAPS_SENSOR=1. Set it to 0 to stop. Calls never wait for 7Maps: outcomes are queued and sent in batches. |
Ready for the LiteLLM proxy, IBM ContextForge, and clients built on the official MCP TypeScript and Python SDKs. MIT licensed, with the exact payload in each README: github.com/XLSV777/7maps.
What we check, and what we do not
7Maps-bot sends only what every MCP client sends first: initialize and tools/list. It never calls a tool. Each tool's risk level (read, change, delete, send, pay) is an automated estimate from its public name, description and annotations, not a security audit, and it can be wrong. Each day's files are hashed, chained to the day before and signed with Ed25519, so anyone can check that history was not rewritten. The full method, with a changelog: methodology.
The world view counts each server in the country where its address is registered, from a public IP database; servers on global edge networks are counted apart. No server's location or address is published. Observations with a date, not a judgement.
Trust and method
| Methodology | What is checked, how often, what each status means, with a changelog. |
|---|---|
| Signed daily history | Every day hashed, chained and signed with Ed25519. Check it in your browser. |
| Verify a badge | Is a 7Maps badge on someone's page genuine and current? |
| How we check, and opting out | 7Maps-bot sends the two standard requests and honours robots.txt. |
7Maps by 7IT, a senior engineering studio. Want a person to review your AI-built app instead? Start here.