The protocol is not the truth

MCP is not a GTM strategy. It is a USB-C port.

RevSure published that on Aug 24. The protocol lets an agent call CRM, ads, and the warehouse without a custom integration per tool. Two MCP servers can expose the same tools and return contradictory answers. One reads a lead source field written at form fill. The other runs attribution across the real touch sequence. Same JSON. Same confident tone. Only one survives a finance review.

Resolve identity once, below the protocol. Then every agent inherits it.

Why does MCP feel like a strategy?

Because teams confuse a clean connection with a clean answer.

MCP (Model Context Protocol) standardizes how an agent asks a system for context and how that system replies. That is useful. It is not truth. A USB-C cable does not decide whether the power supply is honest. It only moves electrons. Same pattern here. The protocol moves context. The graph underneath decides if the context is right.

If you treat MCP as GTM strategy, you will stack servers, celebrate "connected," and still argue in the QBR about which pipeline number is real. Connection is plumbing. Truth is identity, attribution, and write-back rules. That is the same law as the CRM is data, rules, and write-back, not a screen.

What goes wrong when two servers disagree?

You get two confident agents and one burnt AE.

Picture this. Agent A says "this account came from paid LinkedIn last Tuesday." Agent B says "this account has been in nurture for nine months and paid was the last touch." Both answers can look valid inside their own tool. Both can ship as next-best-action text into Slack. The seller opens the meeting with the wrong story. The buyer feels sold, not understood. The protocol did its job. Your identity layer did not.

Asking the model to dedupe three versions of the same account at inference time burns the attention budget on plumbing. That is not AI as operator. That is an expensive intern reconciling spreadsheets in a chat window.

What does the Gartner receipt actually say?

RevSure cites Gartner's May 2026 CSO survey: organizations giving reps AI-enabled next-best-actions were 2.6x more likely to achieve commercial growth.

Read that carefully. The lift is not "we installed MCP." The lift is next-best-actions that reps can trust. If two servers disagree, the next-best-action is noise dressed as AI. You do not get 2.6x from contradictory confidence. You get a faster way to be wrong.

So the receipt supports the point, not the hype. Wire the protocol after the identity graph is resolved. Not before.

How do you resolve identity below the protocol?

Four moves. Do them once. Make every agent inherit them.

1. Pick one source of truth per entity. Account identity. Contact identity. Opportunity stage. Attribution model. Write the owner system on a one-pager. If marketing "owns" lead source and sales "owns" opportunity source with no join key, you already have the RevSure failure mode.

2. Resolve before inference. Dedupe, enrich, and map accounts in the warehouse or CRM rules layer. The agent should read a resolved record. It should not invent which Acme Inc is the real Acme Inc mid-prompt.

3. Expose one graph to MCP, not five raw tools. Prefer a thin GTM context server that already joined CRM, product, and intent. Do not give five agents five conflicting CRM endpoints and hope they reconcile.

4. Govern write-back. Same idea as buy the agent, build the rails. The protocol can offer a write tool. Your rails decide whether that write is allowed. Identity without write rules is a half job.

That stack is GTM engineering: systems that change state under constraints. MCP is the plug. You still own the circuit.

How does this fit Signals?

Signals is who is worth working and why.

If identity is wrong, every signal is wrong. Website de-anon attached to the wrong domain. G2 category traffic scored to a dead account. Competitor LinkedIn engagers merged into a parent that never buys. The protocol will happily ferry those bad joins into every agent that asks.

Ehrenberg-Bass / LinkedIn B2B Institute: about 95% of B2B buyers are not in-market. You cannot afford noisy identity on the 5% who are. Fix the graph once. Then let MCP serve it.

What should I do Monday?

Pick one buyer account that shows up in two systems with different stories. Write which system is source of truth for company name, domain, owner, and last meaningful touch. Stop adding MCP servers until that sentence is boring and documented. Then expose the resolved record to one agent and measure whether next-best-actions match what finance would accept.

Adapt or fail. A clean protocol with dirty identity is still a lie with better JSON.

Start Signals, Convert, Grow

FAQ

Is MCP bad for GTM?

No. MCP is useful plumbing. The failure mode is treating the protocol as truth instead of as a port.

Do I need one MCP server or many?

Many tools can speak MCP. You still want one resolved identity graph underneath. Prefer exposing joined context over raw conflicting endpoints.

How is this different from CRM write-back rules?

Write-back rules decide what an agent may change. Identity resolution decides what the agent is looking at. You need both. Wrong identity plus write access is how bad data spreads faster.

Where does the 2.6x number come from?

RevSure's Aug 24 MCP-for-GTM piece cites Gartner's May 2026 CSO survey: orgs giving reps AI-enabled next-best-actions were 2.6x more likely to achieve commercial growth. That is a next-best-action receipt, not an MCP install receipt.