Agent Toolkit
One place to instrument any autonomous agent with execution assurance: the read-only MCP server, the SDK and framework adapters, the Claude Code skill, and drop-in config for 10+ coding agents — all open source and published today.
The Agent Toolkit for Intentrax gives any AI coding agent the tools to add execution assurance to the software it builds — so an autonomous agent can admit an action, execute it, and produce a verifiable Autonomous Execution Proof. It is the read-only MCP server, the execution-assurance skill, the Claude Code plugin, the SDK and adapters, and drop-in config for the coding agents teams already run. Everything here is open source or published today; only proof-generating execution is review-gated.
Every consequential action produces a cryptographic Autonomous Execution Proof a third party can verify independently — not a log line the operator could have written after the fact.
The public verification API is live today. Verify a real signed proof in one call, with no key and no account, before you wire anything into your own app.
The MCP server exposes exactly two tools and never executes, admits, signs, or writes — so it is safe to hand any coding agent, in any repository.
One MCP command and one config block, documented for 10+ hosts: Claude Code, Cursor, VS Code, Windsurf, Kiro, Codex, Gemini CLI, Qwen Code, and Zed.
The skills marketplace and toolkit hub are public on GitHub; the MCP server and SDK are published on npm. Only proof-generating execution stays review-gated.
Ask your coding agent to check any proof, certificate, passport, or trust asset by identifier and read the signed verdict — without leaving your editor or hand-rolling HTTP.
Wrap a consequential action — a payment, an infrastructure change, a regulated decision — in admit → execute → verify, so the software your agent ships can prove what it did.
Produce a portable, signed proof an auditor, insurer, or counterparty can verify independently, instead of asking them to trust your logs after the fact.
Three surfaces, one toolkit — the server that gives your agent the tools, the skill that teaches it when to use them, and the plugin that installs both.
The read-only server published as @intentrax/mcp-server. It gives any MCP-compatible coding agent two safe tools and never executes, admits, signs, or writes.
The execution-assurance skill teaches an agent when a consequential action is worth a proof, and how to instrument admit → execute → verify correctly.
The Claude Code plugin bundles the skill and the MCP server together, so a single command wires both into your agent.
The command is the same everywhere — npx -y @intentrax/mcp-server — only the config file differs. Public verify needs no key; the tenant tool reads its key and tenant id from your environment. Never commit a key. What works today: intentrax_verify_public_proof is keyless and callable now. intentrax_read_tenant_proof additionally needs an issued tenant and a production API runtime enabled for that tenant, so it is configured but not yet callable — contact us to enable one.
/plugin marketplace add Intentrax/intentrax-skills
/plugin install intentrax-execution-assurance@intentrax[mcp_servers.intentrax-readonly]
command = "npx"
args = ["-y", "@intentrax/mcp-server"]
env_vars = ["INTENTRAX_MCP_API_KEY", "INTENTRAX_MCP_TENANT_ID"]{
"mcpServers": {
"intentrax-readonly": {
"command": "npx",
"args": ["-y", "@intentrax/mcp-server"]
}
}
}{
"mcpServers": {
"intentrax-readonly": {
"command": "npx",
"args": ["-y", "@intentrax/mcp-server"],
"env": {
"INTENTRAX_MCP_API_KEY": "${INTENTRAX_MCP_API_KEY}",
"INTENTRAX_MCP_TENANT_ID": "${INTENTRAX_MCP_TENANT_ID}"
}
}
}
}This site ships no JavaScript by design, so each block is a single copy-ready snippet — select it and copy. Full per-host detail lives on the docs page.
Skills teach your agent when and how to add execution assurance. The execution-assurance skill ships in the marketplace today — it explains when a consequential action is worth a proof and how to instrument admit → execute → verify. Browse or contribute at github.com/Intentrax/intentrax-skills, or install it with the Claude Code plugin above.
Once the MCP server is added, ask your agent:
Verify Intentrax proof <proof-id> and tell me whether the verdict is VALID.
Read the latest execution proof for my tenant and summarize what the agent actually did.
Open the public verifier and check this certificate identifier for me.
| Component | Where to get it |
|---|---|
| Agent Toolkit hub | github.com/Intentrax/agent-toolkit-for-intentrax |
| Skills & Claude Code marketplace | github.com/Intentrax/intentrax-skills |
| Read-only MCP server | npmjs.com/package/@intentrax/mcp-server |
| SDK & adapters | npmjs.com/package/@intentrax/sdk |
| Public verifier | verify.intentrax.com |
The public verification API and the npm packages are free and keyless — verify a real proof with no account. Proof-generating execution is review-gated and unlocks with sandbox access after a short architectural review.
No. The MCP server is read-only: it verifies a public proof and reads a tenant proof, and never executes, admits, signs, or writes. Public verify needs no key and works today; the tenant tool reads INTENTRAX_MCP_API_KEY and INTENTRAX_MCP_TENANT_ID from your environment, and additionally needs an issued tenant plus a production API runtime enabled for that tenant before it can be called. Never commit a key.
Any MCP-compatible host: Claude Code, Cursor, VS Code (Copilot), Windsurf, Kiro, Codex, Gemini CLI, Qwen Code, and Zed. The command is the same everywhere; only the config file differs.
The MCP server gives your agent the verify and read tools. The skill teaches it when and how to add execution assurance. The Claude Code plugin bundles both in one install.
The skills marketplace is at github.com/Intentrax/intentrax-skills and the toolkit hub is at github.com/Intentrax/agent-toolkit-for-intentrax. The packages are @intentrax/mcp-server and @intentrax/sdk on npm.
Developer Path
Start with the no-account action that fits this page; a free account opens the console, and the proof-generating sandbox is issued after a short architectural review.
The Claude Code skills marketplace and the toolkit hub are public repositories under the Intentrax organization — browse, fork, and open issues against the source that powers the toolkit.
The read-only MCP server and the SDK ship on npm today — add the server to any host in one command, or install the SDK to admit, verify, and export proofs.
One MCP command and one config block, documented for Claude Code, Cursor, VS Code, Windsurf, Kiro, Codex, Gemini CLI, Qwen Code, and Zed — plus the one-command Claude Code plugin.
Wrap a consequential agent action in admit → execute → verify with the SDK and adapters, so every action produces a signed, replayable Autonomous Execution Proof.
Start with public developer guidance, then request sandbox access through review.
Get started with Agent Toolkit for IntentraxKeep exploring the public site — no account required.
Docs Quickstart SDKs API Reference Agent Toolkit SandboxResearch, standards, and the AEVN Initiative remain available as public resources.
Explore Intentrax LabsNothing you do here is published. The docs are public to read; your sandbox work stays private to your account, and production access is review-gated.