MCP Layer

01 Core Model

The MCP layer gives autonomous software a native execution surface on TruthLinked. A model client can discover tools, read resources, request prompt templates, and submit agent actions through an MCP-shaped transport. The authoritative records live in cells, storage slots, transaction hashes, and indexed events.

02 Protocol Objects

03 Execution Flow

This flow gives agent systems a stable operational contract. Discovery, authority, execution, and inspection use the same chain-backed state model.

04 Transaction Intents

05 Policy Cell

The policy cell is the control plane for agent authority. It gives the owner a concrete place to define what an agent can read, write, administer, spend, and call.

06 MCP Transport

TruthLinked exposes MCP-compatible HTTP endpoints through the public MCP node gateways. Use mcp1 as the primary endpoint, with mcp2 and mcp3 available as additional node endpoints. The node process also exposes MCP locally on the RPC port plus one.

MCP_URL="https://mcp1.truthlinked.org/mcp"

curl -s "$MCP_URL" \
  -H 'Content-Type: application/json' \
  -d '{"jsonrpc":"2.0","id":1,"method":"initialize","params":{}}'

07 Native Tools

TruthLinked deploys native MCP tools at genesis so agents and clients can inspect the network immediately. These tools provide chain information, balances, validators, staking state, transactions, account history, cells, oracle results, SDK retrieval, faucet guidance, and transaction submission.

MCP_URL="https://mcp1.truthlinked.org/mcp"

curl -s "$MCP_URL" \
  -H 'Content-Type: application/json' \
  -d '{"jsonrpc":"2.0","id":2,"method":"tools/call","params":{"name":"get_chain_info","arguments":{}}}'

08 CLI Usage

The Axiom CLI exposes MCP commands for the registry and execution path. Builders can register agents, resources, prompt templates, and submit agent tool calls from the same operational toolchain used for cells and transactions.

Register an agent

axiom mcp register-agent \
  --from ./owner.keys.json \
  --agent-keyfile ./agent.keys.json \
  --policy-cell-id <policy_cell_id>

Register a resource

axiom mcp register-resource \
  --from ./owner.keys.json \
  --resource-id <resource_cell_id> \
  --name market-feed \
  --uri-scheme trth \
  --mime-type application/json \
  --initial-data-json ./resource-data.json

Register a prompt template

axiom mcp register-prompt \
  --from ./owner.keys.json \
  --prompt-id <prompt_cell_id> \
  --name inspect-cell-risk \
  --template-file ./inspect-cell-risk.prompt.txt \
  --arg cell_id:"Cell ID to inspect":required \
  --arg risk_level:"Low, medium, or high":false

Call a tool as an agent

axiom mcp tool-call \
  --from ./agent.keys.json \
  --tool-id <tool_cell_id> \
  --policy-cell-id <policy_cell_id> \
  --calldata-hex <hex_payload> \
  --gas-limit 500000

09 Prompt Templates

A prompt template is a reusable instruction object with a stable name, argument schema, template hash, and prompt cell ID. It gives agent systems a registry-backed way to reference approved instruction patterns.

MCP_URL="https://mcp1.truthlinked.org/mcp"

curl -s "$MCP_URL" \
  -H 'Content-Type: application/json' \
  -d '{"jsonrpc":"2.0","id":3,"method":"prompts/list","params":{}}'

curl -s "$MCP_URL" \
  -H 'Content-Type: application/json' \
  -d '{"jsonrpc":"2.0","id":4,"method":"prompts/get","params":{"name":"inspect-cell-risk"}}'

A prompt count of zero means the registry currently has zero approved prompt templates visible to the MCP layer and explorer indexer.

10 Explorer Visibility

The explorer exposes MCP activity as a first-class chain surface. Operators can inspect registered agents, tool calls, policy links, resource registrations, prompt templates, and transaction references from the MCP Agents page, transaction pages, address pages, and cell pages.

11 Operating Guidance

• Use tools for callable capabilities with a schema and an authority boundary.
• Use resources for durable data that agents can read and cite.
• Use prompt templates for reusable instruction patterns that need provenance.
• Use policy cells to keep agent authority narrow and reviewable.
• Use action logs and explorer pages to inspect agent behavior after execution.
• Use per-tool permissions for every capability that mutates state or forwards value.