MCP server
@hookforge/mcp gives a coding agent the same view of the catalogue a developer gets from this site:
search by the problem rather than the name, read a hook's parameters, prior art and limitations, resolve its address
on a chain, and decode what any hook address on any chain will actually do.
Add it to a client
{
"mcpServers": {
"hookforge": {"command": "npx", "args": ["-y", "@hookforge/mcp"]}
}
}For Claude Code, one command:
claude mcp add hookforge -- npx -y @hookforge/mcpThe tools
| Tool | What it answers |
|---|---|
search_hooks | "My pool gets arbitraged after every quiet period." Ranked matches, with the terms that matched. |
list_hooks | The whole catalogue, filterable by tag, family or chain. |
get_hook | One hook in full: description, prior art, limitation, parameters, errors, ABI. |
get_deployment | A hook's address on a chain, plus that chain's PoolManager. |
decode_hook_address | Which of the fourteen callbacks any hook address claims, ours or anyone's. |
list_chains | Every chain in the address book with its PoolManager. |
Why an MCP server
An agent asked to add a hook to a pool has two bad options today: guess from a model's training data, which is stale and full of hooks that were never deployed, or scrape a website that was written for humans. Both produce plausible addresses that are wrong.
The server answers from the generated registry, so an agent gets the same facts as the site, including the ones a scrape loses: whether an address is live or merely deterministic, what a parameter's units are, and what the hook explicitly does not do.