When should I pick MCP versus the REST API?

Both reach the same tool catalog and the same customer intelligence layer. Pick MCP when your runtime speaks it natively — the model sees the tool catalog at runtime, picks the right tool, and validates the schema for you. Pick the REST API for deterministic glue code, serverless jobs, or non-MCP agent stacks where the caller already knows which tool to invoke.

What happens when the MCP spec changes?

The Amdahl MCP server is versioned and pinned to a specific spec revision. Breaking changes get a major version bump and a 90-day migration window so your agents stay up without surprise rewrites.

Can I bring my own customer data?

Yes. The MCP server reads from your Amdahl workspace. Whatever you have connected (Gong, HubSpot, Salesforce, Slack, support tickets, docs), the tools can reach. Connector health is visible through the data.explore action.

Does my data leave Amdahl when an agent calls a tool?

Tool responses are sent to the agent runtime you chose (your Claude, your Cursor, your custom client). Amdahl does not send your data anywhere else. Standard MCP semantics. LLM providers queried by Amdahl have zero-retention agreements in place.

What does pricing look like?

Included in any paid Amdahl plan. There is no separate MCP or API SKU. We think every Amdahl customer will be calling us through MCP or the REST API within a year, so we do not gate it.

Is this secure enough for a CTO review?

Yes. SOC 2 Type 2 certified. Per-customer data isolation. TLS 1.2+ everywhere. Scoped, revocable API keys. Rate limits and observability out of the box. No training on customer data, ever.

How does this differ from a CRM integration?

A CRM integration gives an agent access to structured fields (deal stage, contact properties). Amdahl gives an agent access to meaning. The unified layer of customer conversations, pain points, objections, and verbatim language clustered from every source you connected. It is the difference between reading a deal record and actually querying what the buyer said.

The customer truth layer your AI agents run on.

Every call, deal, and signal — normalized into one MCP server and REST API your Claude, ChatGPT, or custom agent can query. Cited answers. Zero retention.

Book a walkthrough

Connect in the Claude UI — no key needed

Claude signs in to your workspace over OAuth. Works in the Claude web app and Claude Desktop — both expose Settings → Connectors.

Open Claude and go to Settings → Connectors (direct link: claude.ai/customize/connectors).
Click Add custom connector.
Name it Amdahl and set the URL to https://app.amdahl.co/mcp.
Click Connect, then Configure.
Sign in to Amdahl when prompted, then choose this workspace.

Need a config-file or CLI path instead? Claude Code, Claude Desktop, and Cursor all speak the same MCP protocol. Generate an API key at app.amdahl.co, then paste the snippet below into the client of your choice. Windsurf, custom Anthropic SDK agents, and custom OpenAI function-calling work the same way.

Claude Desktop

Add to your Claude Desktop config and restart.

json

{
  "mcpServers": {
    "amdahl": {
      "url": "https://app.amdahl.co/mcp",
      "headers": {
        "X-API-Key": "amdhl_YOUR_API_KEY_HERE"
      }
    }
  }
}

Claude Code

One command from the CLI:

bash

claude mcp add --transport http amdahl https://app.amdahl.co/mcp

Cursor

Add to Cursor settings under MCP servers.

json

{
  "mcp": {
    "servers": {
      "amdahl": {
        "url": "https://app.amdahl.co/mcp",
        "headers": { "X-API-Key": "amdhl_YOUR_API_KEY_HERE" }
      }
    }
  }
}

Security and compliance

Built for a CTO review on day one.

Every connection is authenticated, every call is logged, every customer’s data sits in its own isolated tenancy. Same posture across MCP and REST.

SOC 2 Type 2 certified

Audited controls across security, availability, and confidentiality. Per-customer data isolation enforced at the database layer.

Encryption in transit and at rest

TLS 1.2 or higher on every connection. AES-256 at rest. Zero-retention agreements with every LLM provider Amdahl queries on your behalf.

Scoped, revocable API keys

Generate a key per agent, scope it to the tools that agent should reach, revoke it the moment a teammate leaves. Read-only by default.

No training on customer data, ever

Your conversations never enter a model training corpus. The MCP returns answers grounded in your data; the data stays yours.

MCP semantics, plain English

Tool responses are sent to the agent runtime you chose: your Claude, your Cursor, your custom client. Amdahl does not send your data anywhere else. Rate limits, audit logs, and observability ship out of the box.

What agents are doing with this

Four shapes of workflow that builders are wiring up against Amdahl today. Each one is a real production pattern, not a demo script.

AI SDR grounded in verbatim buyer quotes

Every outbound email cites a real customer quote pulled through the data tool. The model never invents a persona.

amdahl.data({ action: "search", query: "ICP pain point: data pipelines", source_type: "transcripts" })

Content agent on a daily cron

A scheduled job calls the assets tool to draft a grounded LinkedIn post every morning. Every claim links to the source call.

amdahl.assets({ action: "generate", query: "today's POV on AI GTM", channel: "linkedin" })

RevOps retention agent

Pulls churn risk signal from the context tool, then drafts personalized retention outreach in the account owner's voice.

amdahl.context({ action: "query_substrate", intent: "answer_question", query: "churn signals this month" })

Battle card drafting from inside Cursor

An engineer working in Cursor asks for a competitor battle card. The MCP returns clustered objection language and verbatim wins, all cited.

amdahl.data({ action: "cluster_search", query: "competitor: ${competitor}" })

Frequently asked

Connect Amdahl to your agent stack in three lines of config.

Book a walkthrough See the product