Integration · Claude Code

Claude Code.

Claude Code is Anthropic's agentic CLI. It runs in your terminal, edits files, runs commands, and reaches outside itself through Model Context Protocol (MCP) servers. Add Agent Cert as an MCP server and Claude Code can certify and verify decisions without ever leaving the terminal — useful when your day-job loop is "agent does work → certify the result → move on."

Add Agent Cert as an MCP server

The fastest way is the claude mcp add command:

claude mcp add --transport http agent-cert \
  https://api.agentcert.net/mcp/sse \
  --header "X-API-Key: ak_live_..."

That writes the config to your user-scope MCP store. Restart Claude Code (or run /mcp in-session and reconnect) and the six Agent Cert tools become discoverable.

Alternative — edit the config file directly

For repeatable / committable setups, drop a .mcp.json in your project root (project scope, shared with the team):

{
  "mcpServers": {
    "agent-cert": {
      "type": "http",
      "url": "https://api.agentcert.net/mcp/sse",
      "headers": {
        "X-API-Key": "ak_live_..."
      }
    }
  }
}

For user scope (across all your projects), put the same config in ~/.claude/settings.json under mcpServers.

Three scopes — pick deliberately

Scope	Lives at	Shared?	Best for
Project	`.mcp.json` in repo root	Yes — checked into git	Team repos; ensures everyone gets the same MCP config
User	`~/.claude/settings.json`	No — your machine only	Your personal setup across all projects
Local	Project-specific, not in git	No	One-off project testing

For Agent Cert: user scope is usually right — your API key is personal. Project scope works if you want the team to share the connection but each developer brings their own key (use ${ENV_VAR} substitution in .mcp.json).

Keep the API key out of git

If you go with project scope, don't commit the key. Use env-var substitution:

{
  "mcpServers": {
    "agent-cert": {
      "type": "http",
      "url": "https://api.agentcert.net/mcp/sse",
      "headers": {
        "X-API-Key": "${AGENT_CERT_API_KEY}"
      }
    }
  }
}

Each developer exports AGENT_CERT_API_KEY in their shell and Claude Code resolves it at connection time. Same pattern works in ~/.claude/settings.json.

Verify the connection

In Claude Code, run:

/mcp

You should see agent-cert listed as connected, with all six tools (certify_decision, get_receipt, verify_receipt, check_credits, list_my_receipts, get_anchor_proof) available.

Then ask: "Check my Agent Cert credit balance." If you get a number back, you're live.

What you can do once connected

Certify a deliverable. After Claude Code completes a non-trivial task (a refactor, a migration, a deploy decision), ask it to certify_decision with the task summary as the agent_output. You get a signed, anchored receipt — useful when the change touches regulated or audit-relevant code.
Verify a receipt. Drop a receipt ID in a chat and ask Claude to verify_receipt — handy when reviewing a customer's claim or auditing an old decision.
Browse recent receipts. list_my_receipts pulls your last 25 receipts. Useful for "what did my agents decide this week?" review sessions.
Develop integrations. Build and test your own Agent Cert clients from inside Claude Code — the CLI can hit the API directly and tell you what shape the responses are.

Tool Search note

Claude Code uses Tool Search — instead of loading every connected tool upfront, it discovers them on-demand based on the conversation. So Agent Cert tools won't inflate your context window unless you ask for them. No performance trade-off for adding the connection.

Costs

Each certify_decision call: 3 Agent Cert credits.
Other tools (get_receipt, verify_receipt, check_credits, list_my_receipts, get_anchor_proof): free — read-only.