MCP server + agent skill for gno.land.
gnomcp connects gno.land to any MCP client (Claude Code, Claude Desktop, Cursor, Gemini CLI, OpenCode, …): read realms, evaluate expressions, inspect accounts, manage testnet keys, and simulate or broadcast transactions.
- MCP server — the tools to read and write gno.land from your agent, with safety built in.
gnoskill — the knowledge layer for coding agents: interrealm semantics, security taxonomy, idiomatic patterns,Render()conventions, stdlib surface.
Warning
Work in progress — unaudited and pre-release.
- The tool API can still change, and the session write path will be reworked.
- Writes are confined to dev/testnet — no code path signs on mainnet or betanet.
- No guaranteed upgrade path yet.
Read docs/security.md and file issues when something looks off.
One command installs everything — the server binary and the plugin (skills + agents). It downloads the binary into ~/.local/bin (verifying the checksum) and wires up the clients it can: Claude Code and Gemini CLI automatically, Codex and OpenCode with printed steps.
curl -fsSL https://raw.githubusercontent.com/gnoverse/gno-mcp/main/scripts/install.sh | shThis runs a script from the internet on your machine — read the script first.
When it finishes, restart your editor or agent so it loads gnomcp.
Other clients (Cursor, Claude Desktop, …), manual install, building from source, and Docker → docs/gnomcp.md.
gnomcp ships pointed at the public testnet and a local gnodev node — nothing to configure:
| Profile | Chain-id | RPC |
|---|---|---|
testnet |
test-13 |
https://rpc.test13.testnets.gno.land:443 |
local |
dev |
http://127.0.0.1:26657 (local gnodev node) |
Then talk to your agent in plain language. New to gno.land? Just ask it to teach you — it gauges your background and gives you a hands-on tour (the gno-onboard skill). Otherwise:
"Which chain is this, is the node live, and what realms exist under gno.land/r/test? Show me what the counter realm renders."
Reads live state — node status, the realm catalog, account balances, a realm's source and rendered page — grounded in real queries, never guessed. Works immediately, on any chain.
"Deploy a check-in board at gno.land/r/test/checkin: anyone can check in, it records their address, and reading it back lists everyone so far. Then check in yourself."
Your agent writes the realm, tests it locally, runs a security pass, then — once you pick where it runs — funds a key from the faucet, deploys, and makes a real call to prove it works. You get working on-chain code and the transaction that proves it.
"Give me a formal security audit of gno.land/r/test/vault before I route user data through it."
Fetches the on-chain source (read-only — works on mainnet too) and returns an evidence-backed report: findings with quoted lines and severity, plus an honest note on what it did and didn't check. Nothing is mutated.
"My transaction failed with insufficient_funds — what happened?"
Classifies the error, reproduces it cheaply without broadcasting, applies the fix, and re-runs to prove it works — telling you which identity signed each attempt.
"From now on write as my account g1… , not your own key. Bump the counter at gno.land/r/test/counter as me."
Your agent proposes a scoped session and hands you a gnokey command to authorize on your own machine — it never touches your keys. Once you approve, it writes as you, within the limits you set, and tells you exactly how to revoke.
Under all of these, the gno skill gives your agent the language, idioms, and security model; the workflows above build on it. Skill authoring → docs/skills.md.
23 tools, grouped by what they touch:
- Chain reads — render realms, evaluate expressions, read packages, inspect accounts and status. Work immediately.
- Indexer reads — list realms, deploy and transaction history, on-chain activity. Need a profile with an indexer URL.
- Writes — call functions, run code, deploy packages. Need a funded agent key or an active session.
- Sessions & keys — propose and revoke user sessions; generate, list, fund, delete agent keys, and transfer ugnot between a profile's own keys.
Full catalog → docs/tools.md.
gnomcp can reach any gno.land chain. Dev and testnet chains are read/write; mainnet and betanet are read-only — inspect and audit deployed code, but no signing on real-funds chains.
Beyond the built-in testnet and local defaults, save the chains you use as named profiles with gnomcp profile add (written to profiles.toml), so gnomcp remembers them between runs. A profile can also carry an indexer URL, or a master address for user-session writes (dev/testnet only).
Profile fields and the signing model → Configuration · Write authorization.
- Keys stay in
gnokey— gnomcp never sees a mnemonic; the user signs sessions on their own machine. - No signing on real-funds chains — writes are gated to dev/testnet; mainnet and betanet are read-only.
- Chain output can't hijack the agent — every chain-derived byte is wrapped in an untrusted-content envelope.
- Bounded reads — output is budgeted and summarized, never silently truncated.
- Every write is logged — an append-only audit trail of the tool, profile, result, and signer.
- Structured errors — machine-routable codes and recovery hints, so the agent fails forward.
Full posture and threat model → docs/security.md.
make test # Unit tests (no network)
make test-integration # In-process node + live tests (build tag: integration)
make lint # go vet + gofmt -l
make build # bin/gnomcp
make dev # go run ./cmd/gnomcp (starts MCP server)Built against the official MCP Go SDK (github.com/modelcontextprotocol/go-sdk) and github.com/gnolang/gno/gno.land/pkg/gnoclient for chain RPC. Testing: see test/README.md for the four test layers (unit / integration / agent e2e / manual).
- Thin the skills toward the gnolang/gno monorepo as the single source of truth, less hand-distilled content
- Docker as the canonical deployment
- a2a serve mode (agent-to-agent realm bridge)
- External security audit before any "stable" claim
Apache-2.0. See LICENSE.