Pensero MCP lets you query your Pensero data directly from Claude, ChatGPT, Gemini or another MCP-compatible AI assistant. Instead of navigating through different views or building a report, you ask a question in plain language and continue with follow-up questions in the same conversation.

The assistant does the thinking. Pensero provides the engineering data it thinks with.

MCP (Model Context Protocol) is an open standard that lets AI assistants (like Claude or ChatGPT) securely connect to external tools and data sources and use them in a conversation. Instead of copying data out of Pensero and pasting it into a chat, MCP lets your AI assistant ask Pensero directly for the numbers it needs, using your existing Pensero permissions.

Pensero MCP is Pensero's implementation of this standard. Once connected, your assistant gets access to the same Pensero data you already use for delivery, quality, code review activity, AI impact, collaboration and benchmarks, and pulls real, live numbers from Pensero to answer.

Some examples:

You can also ask follow-up questions without starting again, compare people or teams, and dig deeper into any answer, the same way you would in a normal conversation. For example:

For a longer list by theme, see Pensero MCP prompts below.

You are not limited to views that already exist in Pensero. When the question you want to answer doesn't fit an existing dashboard, describe the comparison you need, select a group of people or period, and ask the assistant to analyze it. For example:

You describe what you want to understand, and your assistant queries Pensero for the relevant data.

If your assistant is also connected to tools such as Jira, Linear, your CRM, or incident management tools, you can combine them in the same question. Pensero provides the engineering performance and AI impact data. The other system can provide the roadmap, incidents, sales data, or other context. For example:

Pensero MCP is available to everyone in your organization: executives, managers and individual contributors. Your organization admin can turn access on or off for each role under Organization settings → MCP access.

What the assistant can see depends on your role, the same way it does in the Pensero web app:

You'll need:

The Pensero MCP server address is:

Tip: Sign in to pensero.ai in your browser before you connect. If you're not signed in, the connection flow opens Google sign-in. If you normally sign in to Pensero with GitHub, Microsoft or single sign-on (SSO), sign in first and the connection will use that session.

Pro and Max plans

Team and Enterprise plans

A Claude Owner first adds Pensero for the whole organization: Organization settings → Connectors → Add → Custom → Web, enter https://pensero.ai/api/v1/mcp, then Add. After that, each person goes to Customize → Connectors, finds Pensero (labelled Custom) and clicks Connect.

In ChatGPT, custom MCP connections are called apps. They need developer mode and are available on the Business, Enterprise and Edu plans, on ChatGPT web.

To share it with the rest of your ChatGPT workspace, a workspace admin publishes it from Workspace settings → Apps → Drafts → Publish.

Gemini Enterprise asks you to fill in the sign-in details by hand. A Gemini Enterprise admin adds Pensero as a custom MCP server and enters these values:

Save, then click Connect and sign in with your Pensero account. You'll be sent back to Gemini.

This Client ID works with the Gemini redirect URI https://vertexaisearch.cloud.google.com/oauth-redirect. If Gemini shows a different Redirect URI, or you get the error redirect_uri not registered, send us that Redirect URI in the Pensero chat and we'll give you the right Client ID.

Any assistant that supports remote MCP servers with OAuth sign-in can connect. Add a new MCP server with the address https://pensero.ai/api/v1/mcp and sign in to Pensero when asked. Most assistants don't need a client ID or secret: they register themselves automatically. If your assistant asks for an Authorization URL, Token URL and Client ID, use the values in the Gemini Enterprise section above and send us its Redirect URI in the Pensero chat so we can give you the right Client ID.

In Pensero, go to Data health → Link more accounts → Pensero MCP to see the connection steps. Once connected, each assistant appears in your linked accounts, where you can see its recent activity and click Revoke to disconnect it. Revoking stops that assistant from reaching your Pensero data straight away; to use it again, connect it again.

Once Pensero MCP is connected, you do not need to retype a question. Anywhere Pensero shows a metric or a signal, an Ask AI control opens that question in your assistant, already written for you.

Where you will find it

What gets sent

The question carries the context of the page you are looking at: the metric, the period selected, and the scope (the person, team, or cohort selected in the global selector). Your assistant then queries Pensero through MCP for that same window and those same people, so the answer matches the screen you started from. From there you keep the conversation going with follow-up questions.

Choosing your assistant

The first time you click Ask AI, Pensero asks which assistant you use: Claude, ChatGPT, Gemini, or Grok. Keep "Save as my preference" enabled and every later Ask AI click goes straight there, with that assistant’s logo on the control. You can change it at any time in Account settings > AI assistant.

Requirements

These are starting points. Pensero's MCP exposes the full depth of your engineering data. The prompts below are organized by theme and user type, with a deep-dive investigation at the end.

"How much did engineering ship in the last 30 days, and what share was human vs AI-assisted vs agentic?"

"Who were the top 5 contributors on the backend team last quarter?"

"Give me the exec-level engineering health snapshot — our core signals with trends and how we compare to the industry."

"What percentile is Alice in for delivery compared to her level peers?"

"What's our defect cost this quarter, and who introduced the most rework?"

"Is our rework rate trending up over the last 90 days?"

"How does Alice's defect rate compare to the backend team's?"

"What's our P90 time to merge, and which repositories are the slowest?"

"Show me the 20 slowest merged PRs and the tickets driving them."

"Rank the engineering team by cycle time, slowest first."

"Who reviews the most, and how good is their review quality?"

"Show me recent code review activity on the platform team."

"What's our AI adoption — AI code %, cost per delivery point ($/pt), tokens per unit of delivery, and total AI spend?"

"Who uses AI most efficiently on the engineering team — best $/point and tokens per point?"

"Break down our AI spend by tool and by model."

"How much did AI agents deliver versus humans, at what cost per point, and who invoked them?"

"Show me this month's agent sessions with their linked PRs and usage."

"Give me a full performance profile for Bob for 1:1 prep."

"Summarize the platform team: delivery, top contributors, and weekly trend."

"Compare full-time employees with over a year of tenure against the rest of the org on delivery and quality."

"What attributes can I use to filter people into a group?"

"What is the team working on — top roadmap epics versus off-roadmap work?"

"Break down last quarter's work by category: features, bugs, and KTLO."

"What's our capitalizable work percentage and cost per point for finance reporting?"

"How do we stack up against the industry on cycle time and defect cost?"

"What percentage of completed work had score edits last month, and were the downward adjustments justified?"

"How much thumbs-up/down feedback are people giving on work-item scores, and who isn't engaging?"

"Are all our integrations healthy? Which need attention?"

"Show me the last 10 failed syncs for our GitHub org."

"Which GitHub account is linked to Alice, and what's her team and manager?"

"Count merged PRs since June, grouped by author, ordered by most PRs."

"What data can I query?"

A four-prompt flow you run in sequence to understand why delivery is low across the org — and whether AI adoption or annotations explain it.

Prompt 1 — Delivery distribution across the org

Pull delivery for everyone in my org for the last 30 days. Bucket everyone by delivery, using the org's own percentiles as thresholds: Zero (delivery = 0), Very low / low (above 0, up to P25), Normal lower (P25 to P50), Normal upper (P50 to P75), High (above P75). For each bucket show headcount and % of total, total delivery points, and average delivery per head. Also show me the actual P25, P50 and P75 values you used as thresholds.

Prompt 2 — Do annotations explain the low delivery, or just cover it?

Now pull annotations for the same people and period. Per person, give me the total number of calendar days covered by annotations (not the count of annotations). Sort each person's annotations into two buckets by reason: Reasonable (legitimately explains low delivery: leave, sick days, onboarding, company holiday, internal project with no code output, offboarding transition) and Not reasonable (vague, missing, or unconvincing: "working on tasks", "busy", no reason given, or anything that doesn't explain absent delivery). Then, for each delivery bucket from Prompt 1 and separately for each team, show: how many people have at least one annotation, total annotated days split reasonable vs not reasonable, average annotated days per person in that group, and average delivery of annotated vs non-annotated people in the same group. The goal is to see whether annotations genuinely explain low delivery or just paper over it — and whether that pattern differs by team.

Prompt 3 — AI adoption across delivery buckets

For the same org and period, pull AI adoption metrics per person. For each delivery bucket (Zero / Very low / Normal lower / Normal upper / High) show: % of people in the bucket who used AI tools at least once (AI adoption), average AI code % (share of merged code that was AI-assisted), and average delivery per head split by human-only vs AI-assisted contributors. The goal is to see whether AI adoption correlates with which delivery bucket someone lands in.

Prompt 4 — Teams vs the whole org on AI adoption

Now compare each team against the full org on AI adoption. For each team show: AI adoption % (team vs org), AI code % (team vs org), average delivery per head (team vs org), and which delivery bucket the team's median contributor falls into. Flag teams above org-average AI adoption but below org-average delivery — where AI usage may not be translating into output. And the inverse — high delivery, low AI adoption — as potential headroom.

Does this give the AI assistant access to everything in Pensero?

No. It only sees what you can already see in Pensero: executives see the whole organization, managers see the people who report to them, and individual contributors see only their own data.

Is my data sent anywhere outside Pensero?

Your AI assistant sends your question to Pensero, and Pensero returns the relevant data directly to that conversation. No separate copy of your data is stored by Pensero MCP beyond normal request logging used for reliability and support.

Can I disconnect at any time?

Yes. Go to your linked accounts in Data health and click Revoke next to the assistant.

Why can’t I access Pensero MCP?

Your organization admin decides which roles can use MCP. Ask them to check Organization settings → MCP access and turn it on for your role.

The connection worked, but my assistant says "MCP API is not available for your account".

MCP access is turned off for your role. The connection can still finish, but every question will fail until an organization admin turns it on for your role under Organization settings → MCP access. You don't need to reconnect afterwards.

I was sent to Google sign-in but I don't use Google to sign in to Pensero.

Close that window, sign in to pensero.ai the way you normally do, then start the connection again from your assistant.

Gemini Enterprise says "redirect_uri not registered".

The Client ID doesn't match the Redirect URI that Gemini uses. Send us the Redirect URI that Gemini shows in the Pensero chat and we'll give you the right Client ID.