ConexiConexi: Where Intelligence Meets Integration.
Conexi connects AI agents to tools, data and chains without handing them everything. Set up MCP connectors, decide per agent which tools it may call, make sensitive calls wait for your click, and keep a record of every call. Connecting AI, tools and data for the onchain era.
Sensitive tools always stop for a person, whatever the allow list says.
LiveFresh from Robinhood Chain blocks and the Chainlink ETH / USD feed · policy registry not deployed yet
How it works
A few minutes and no account, in three moves. Your agent keeps its own client; the hub gives it a smaller, safer world to work in.
Pick the connectors
Each connector lists the tools it usually exposes, already sorted into read, write and sensitive.
Write the policy
{
"mcpServers": {
"fetch": {
"command": "uvx",
"args": [
"mcp-server-fetch"
]
},
"filesystem": {
"command": "npx",
"args": [
"-y",
"@modelcontextprotocol/server-filesystem",
"/Users/you/projects/notes"
]
}
}
}Allow, deny and approval rules per agent, with wildcards and a call limit. Default is deny.
Route and approve
Every call lands in the activity monitor, refused and rejected ones included.
Tools that really run
12 built-in tools for the routing playground. Read tools answer anyone the policy allows; sensitive ones change stored data and always wait for a person. Tools you register from other connectors are routed and logged too, then handed to your own client.
| Tool | Purpose | Access |
|---|---|---|
| chain_get_block | Latest Robinhood Chain block: number, time, transactions, gas used. | Read-only |
| chain_gas_price | Current gas price on Robinhood Chain, in gwei. | Read-only |
| chain_eth_usd | ETH / USD from the Chainlink feed on Robinhood Chain, with its update time. | Read-only |
| chain_get_balance | ETH balance and transaction count of an address. | Read-only |
| chain_token_info | Name, symbol, decimals and total supply of an ERC-20 token, read from its contract. | Read-only |
| chain_get_tx | Status, sender, recipient, value and block of a transaction. | Read-only |
| hash_text | SHA-256 or keccak-256 of a text, computed in your browser. | Read-only |
| json_validate | Checks a JSON value against a JSON Schema and lists every problem. | Read-only |
| json_pick | Reads one value out of a JSON document by a dotted path such as items.0.name. | Read-only |
| notes_read | Reads the playground notebook kept in this browser. | Read-only |
| notes_write | Adds a line to the playground notebook in this browser. Marked sensitive: it changes stored data. | Needs you |
| notes_clear | Erases the playground notebook in this browser. Marked sensitive: it deletes data. | Needs you |
Your policy decides.
Risky calls wait.
The router reads the policy before anything runs: deny rules first, then the allow list, the call limit and the input schema. A tool marked sensitive then stops and waits in your approvals queue. The agent can ask; it cannot skip the line.
| Action | Agent | You |
|---|---|---|
| Call a tool on its allow list | ✓ | ✓ |
| Call a tool on its deny list | ✕ | ✓ |
| Call a tool nobody listed | ✕ | ✓ |
| Run a sensitive tool without a click | ✕ | ✓ |
| Go past its call limit | ✕ | — |
| Send arguments the schema refuses | ✕ | ✕ |
| Change its own policy | ✕ | ✓ |
| Approve or reject a waiting call | ✕ | ✓ |
| Publish the policy fingerprint | ✕ | ✓ |
| Plain answers | |
|---|---|
| Who decides | Your policy, checked on every call |
| Default | Deny: an agent starts with no tools |
| Sensitive tools | Wait for a person, every time |
| Secrets | Never typed here; placeholders only |
| Hub data | Kept in your browser |
| On chain | A name and two hashes, nothing else |
| Contract owner | None. No admin, no proxy, no fee |
| Registry | ConexiPolicyRegistry, not deployed yet |
The worst case of a leaked agent setup is calls inside its allow list until you change the policy. Read the security model → · chain 4663
What's live, what's planned
6 of 10 pieces work today, in your browser and on Robinhood Chain. The rest is planned and marked that way. $CONEXI is the project token; its utility belongs to the planned part.
| # | Piece | Share of the plan | Status |
|---|---|---|---|
| 01 | MCP Connector HubTemplates for tool servers, client config file | Live | |
| 02 | Tool RegistryTools with JSON Schema checks and access levels | Live | |
| 03 | Permission ControlsAllow, deny, approval and limits per agent | Live | |
| 04 | Tool Call RoutingPlayground where built-in tools really run | Live | |
| 05 | Human ApprovalSensitive calls wait for your click | Live | |
| 06 | Activity MonitorEvery call with status, kept in your browser | Live | |
| 07 | Policy RegistryContract ready and tested; practice mode until the owner deploys it | Practice | |
| 08 | Developer APIConnectivity for AI apps over an API | Planned | |
| 09 | Hosted connectorsConnectors that run for you, not on your machine | Planned | |
| 10 | $CONEXI utilityService credits, integration capacity, premium monitoring | Planned |
$CONEXI · at launch on PonsPons launchpad ↗Token page →
Ask Patch.
What people ask before they let an agent near their tools. Pick a question and Patch answers.
01you › What is Conexi?
Conexi is a connection layer between AI agents and the tools they use through the Model Context Protocol (MCP). It does not build another agent. It helps you set up connectors, list the tools each one exposes, decide which agent may call which tool, and see every call that went through.
02you › How does Conexi work today?
Everything in the hub runs in your browser. You pick connector templates and get a client config file, register tools with a JSON Schema for their input, write a permission policy per agent, and try calls in the routing playground. Built-in tools really run: Robinhood Chain reads, hashing, JSON checks and a small notebook. Sensitive calls stop and wait for your approval.
03you › Can an agent call a tool I did not allow?
Not through the router. Every agent starts with no tools. A call goes through only when an allow rule names the tool and no deny rule does, its arguments match the tool's schema, and it is under the agent's call limit. Sensitive tools wait for a human click on top of that.
04you › Does Conexi see my keys or files?
No. Connector templates write placeholders where secrets go, so you fill them in your own client, never on this site. The manifest you can publish leaves out folders, URLs and keys entirely. The hub's data stays in this browser's storage.
05you › What goes on Robinhood Chain?
Only fingerprints. ConexiPolicyRegistry stores a name, the keccak-256 hash of your connector manifest and the hash of your policy, plus who published them and when. Anyone holding the files can recompute both hashes and check them against a release. The contract has no owner, no admin and no fee beyond gas.
06you › Which MCP clients work with the generated config?
The config uses the common mcpServers layout most desktop and editor clients read: a command and arguments for local servers, a URL for remote ones. Check your client's documentation for where the file goes. The templates point at well-known reference servers and at your own.
07you › Is the AI part required?
No. The playground reads plain requests with a transparent rule set and labels it that way. If the site operator adds an AI key, a model may propose the tool call instead; the router still checks the policy, the schema and the approval step, exactly as before.
08you › How do I buy $CONEXI?
$CONEXI has not launched yet. When it does, it will trade on Pons, the Robinhood Chain launchpad, and the contract address will appear in the navbar, the footer and on the token page. Until then, any address claiming to be $CONEXI is not ours.
09you › What is $CONEXI for?
The plan is service credits, more integration capacity and premium monitoring once hosted parts exist. None of that is live, so $CONEXI gives no access, discount or return today, and the hub works fully without it.
10you › What are the risks?
The contract is young and has not been through an independent review; it stores hashes only, so the funds at stake are your gas. Launchpad tokens swing hard and may lose all their value. A policy in this hub is a plan for your agents; your MCP client and servers still decide what actually runs on your machine.
