Now with MCP tool-server support

The AI command center for executives, not chatbots.

Outcome-driven recipes, your own AI provider, live business data via MCP, and total privacy control. Connect Claude, OpenAI, OpenRouter, ZenMux, or a fully local Ollama model.

No markup on AI usage — you pay providers directly. Your keys never leave your device.
💬 Ask ExecPilot Anything MCP Connected
"What were my earnings last quarter?" You
🔧 Calling: reports__get_earnings Q3-2025
Q3 revenue: $4.2M (+12% QoQ). Full report ready — ⬇ Download .xlsx
Bring your own model
Claude
OpenAI
OpenRouter
ZenMux
Ollama (Local)

Built for outcomes, not conversations

Every recipe is a structured workflow — inputs, steps, and a finished output — so you get answers, not another blank chat box.

🧩

Recipe Packs

Pre-built workflows for CEOs, CFOs, CTOs, CHROs and CMOs — board deck reviews, contract risk scans, investor emails, and more.

🔧

Live Data via MCP

Connect internal tool servers over the Model Context Protocol. Ask a question — the AI routes it to the right tool and returns real numbers, not guesses.

🔌

Any Model, One Interface

Connect Claude, OpenAI, OpenRouter, ZenMux, or a local Ollama model. Switch providers per recipe — no lock-in, ever.

🔒

Local-First Privacy

Sensitive material — HR, M&A, financials — can be routed to run entirely offline on your machine via Ollama. Tool-calling works locally too.

⚙️

Code-Execution Skills

For spreadsheets, the AI writes a transformation script that runs locally on your full dataset — exact math, any file size.

🛠️

Your Own Recipes & Skills

Build custom workflows in a visual editor. Save reusable skills — your voice, your company context, your rubrics.

Ask a question. Get real numbers from your own systems.

ExecPilot speaks the Model Context Protocol (MCP) — the open standard for connecting AI to tools. Point it at your internal reporting server and the AI stops guessing and starts querying.

  • Connect multiple serversHTTP or SSE transport, with bearer-token or custom-header auth. Your finance server, your CRM connector, your BI gateway — all at once.
  • Automatic tool routingAsk in plain English. The AI discovers your server's tools, picks the right one, fills in the arguments, and calls it — with a visible trace of every call.
  • Reports come back as ExcelWhen a tool returns tabular data, ExecPilot detects it and offers a one-click .xlsx download.
  • Permission-gated by designExecPilot only gains access to a server's exact origin when you explicitly approve it — Chrome prompts you, per server.
💬
You ask

"What were my earnings last quarter?"

🧠
AI picks the tool

Discovers get_earnings_report on your Finance MCP server and calls it with period: Q3-2025.

🔧
Server responds with real data

Your own system returns the rows — the AI never invents a number.

📊
Answer + Excel download

A plain-English summary on screen, and the full report one click away as .xlsx.

From input to outcome in four steps

No prompt engineering required. Pick a recipe, add your content, get a finished result.

1

Connect your model

Add an API key or point to a local Ollama instance. Keys stay on your device.

2

Pick a recipe

Choose from built-in packs, a custom recipe, or just Ask ExecPilot Anything.

3

Add your content

Paste text, capture a selection, drop in a file — or let MCP tools fetch live data.

4

Get your outcome

A finished brief, email, table, or downloadable Excel file — ready to use.

A recipe pack for every seat at the table

Switch tabs to preview what's inside each role-specific pack.

📋
Board Deck Reviewer

Finds red flags, gaps, and the hardest questions a director will ask.

🔭
News → Strategic Implications

Turns market or competitor news into a threat/opportunity brief.

✉️
Investor Update Drafter

Drafts investor emails in your saved voice and tone.

⚖️
Steelman the Opposition

Stress-tests a decision with the strongest opposing case.

📊
Financial Report Summarizer

Upload a PDF or XLSX for an instant variance analysis.

📑
Contract Risk Scanner

Flags liability, indemnity, and renewal clauses that need counsel review.

🔧
Live Earnings Queries (MCP)

Ask for last quarter's numbers — pulled straight from your reporting server.

⚙️
Spreadsheet Transformer

Cleans, merges, and recalculates spreadsheets — returns a real .xlsx.

🔐
Vendor Security Review

Analyzes a vendor's security questionnaire for gaps.

⚖️
Build vs. Buy Scorecard

Evaluates a technical proposal against your criteria.

🧯
Incident Postmortem Summarizer

Turns a raw incident log into a clean postmortem doc.

🗺️
Architecture Doc Reviewer

Flags scalability and security concerns in design docs.

🔴
Sensitive Comms Drafter

Local-model-only by default — for HR matters that can't touch the cloud.

📄
Policy Document Comparator

Diffs two policy versions and summarizes material changes.

💬
Engagement Survey Themes

Extracts themes and sentiment from open-text survey responses.

📋
Job Description Generator

Drafts role descriptions consistent with your leveling framework.

💬
Ask ExecPilot Anything

Free-form requests, with live MCP tool-calling when servers are connected.

📋
Executive Brief

Turns anything into 3 bullets plus a recommendation.

Notes → Decisions & Actions

Extracts owners and deadlines from raw meeting notes.

✉️
Draft Email Reply

Summarizes a thread and drafts your reply in seconds.

ExecPilot Pro

From one-off answers to an auditable operating system

Chain work, enforce output standards, compare models, verify evidence, and keep a decision trail—all from the Chrome side panel.

⛓️

Recipe chains

Connect compatible recipes so each result becomes the next step’s input, with optional approval checkpoints.

Use it: Pro Tools → name workflow → add steps → Run
🔎

Evidence-backed outputs

Ask the model to ground claims in numbered source passages, then click citations to inspect the supporting text.

Use it: open recipe → enable Evidence-backed output
↔️

Spreadsheet change review

Inspect changed cells, added rows, and removed rows before approving the transformed workbook for download.

Use it: Code Transformer → run → review → approve

Output contracts

Define required sections or fields. ExecPilot validates the answer and makes one correction attempt when something is missing.

Use it: + New → Output contract → Save Recipe
📌

Decision workspace

Save recommendations with alternatives, assumptions, risks, evidence, owner, deadline, and follow-up date.

Use it: run recipe → Save decision → Pro Tools
🛡️

Human approval gates

Consequential MCP actions such as sending, updating, deleting, publishing, or paying pause for explicit approval.

Use it: approve the server, tool, and arguments shown
⚖️

Model comparison

Run the same recipe and input through two configured providers and inspect the results side by side.

Use it: Pro Tools → Model comparison
🧪

Recipe versions and tests

Save prior recipe versions, restore them later, and test drafts against representative inputs and expected terms.

Use it: + New → Test input → Test draft
🏢

Company context packs

Maintain reusable terminology, strategy, products, KPIs, and policy context, then apply it only when relevant.

Use it: Pro Tools → Company context → select in recipe
📊

Themed table output

Detected CSV, Markdown tables, and structured spreadsheet results render as readable tables with safe Excel downloads.

Use it: request a table or run a spreadsheet recipe
📄

Executive export studio

Turn the current result or a saved decision into a branded board memo, decision paper, or presentation outline.

Use it: run recipe → Export executive report
🔐

Your provider, your control

Use Claude, OpenAI, OpenRouter, ZenMux, or local Ollama while keeping provider credentials on your device.

Use it: Providers → configure → Save
Example workflowClean spreadsheetReview changesAnalyze varianceVerify evidenceSave decisionExport board memo

Powerful automation, deliberate control

Approval gates, output contracts, evidence links, and before/after review keep people in charge of consequential work.

Explore Pro
DEFENCE IN DEPTH · ON-DEVICE ENFORCEMENT

Untrusted content stays data—not authority.

ExecPilot adds host-level controls around document Q&A and MCP tool use, reducing the chance that instructions hidden in a file or tool response can redirect the model or trigger an unintended action.

🛡️

Injection screening

Extracted document passages, MCP definitions, and tool results are checked locally for instruction overrides, role impersonation, secret extraction, encoded payloads, and external-send patterns.

🚧

Quarantined evidence

High-risk document passages and MCP results are withheld from the answering model. Users see a clear warning and the reason content was constrained.

🎯

Least-privilege tools

No MCP tool is exposed automatically. Users select the exact servers and tools permitted for each run, keeping the model's action surface intentionally narrow.

Exact-call approval

Every MCP invocation pauses for confirmation showing the server, tool, risk context, and exact arguments. A changed call requires a new approval.

🔎

Schema and size enforcement

Tool arguments are validated against their declared JSON schema. Definitions, arguments, responses, and model-bound results are bounded to limit malformed or oversized payloads.

🔒

Tainted-session controls

Once untrusted MCP output enters a run, subsequent proposals carry an elevated warning. Tool traces render as inert text so server-controlled markup cannot execute in the extension UI.

Designed for enterprise review: security controls run in the extension, require no additional Chrome permission, and are available regardless of plan. Prompt injection cannot be eliminated by any LLM product, so ExecPilot combines model instructions with deterministic screening, least privilege, validation, quarantine, and human approval.

Privacy isn't a setting. It's the architecture.

  • No middleman serverEvery call goes direct from your device to the provider or MCP server you chose. ExecPilot never sees your prompts.
  • Local-only modeRoute sensitive recipes to Ollama running on your own machine — nothing ever reaches the internet.
  • Per-origin permissionsMCP servers get access only after you approve their exact origin in a Chrome permission prompt — revocable anytime.
  • Keys stored locallyBrowser-local storage in Chrome, DPAPI-encrypted on Windows. Never logged, never exported.
Read the full Privacy Policy →
🔴 Local Tier

Ollama only. Zero network egress. HR, M&A, board materials.

🟡 Approved Cloud

Pre-vetted providers for internal, non-public material.

🟢 General Cloud

Any connected provider. Fast, low-cost, for public or low-sensitivity tasks.

Setting up local Ollama (fixing CORS)

Browsers block extensions from calling local services unless the service allows it. Ollama needs one environment variable — OLLAMA_ORIGINS — to accept requests from ExecPilot. Pick the option that suits you.

Sets the variable only for the current PowerShell window. Ollama accepts extension requests until you close that window or stop the server. Good for trying things out.

# Stop Ollama if it's already running (quit it from the system tray first), then: $env:OLLAMA_ORIGINS = "chrome-extension://*" ollama serve

Keep this PowerShell window open while you use ExecPilot. When you close it, the setting is gone.

Sets the variable permanently for your Windows account. Survives reboots. No admin rights needed. Recommended for most people.

# Run in any PowerShell window (no admin needed): [System.Environment]::SetEnvironmentVariable("OLLAMA_ORIGINS", "chrome-extension://*", "User")

Then fully restart Ollama: right-click the Ollama icon in the system tray → Quit, and start it again (or reboot). New environment variables are only picked up on a fresh start.

Sets the variable system-wide for every user account on the machine. Requires an elevated (Administrator) PowerShell. Use this on shared workstations or when Ollama runs under a service account.

# Run PowerShell AS ADMINISTRATOR (right-click → Run as administrator): [System.Environment]::SetEnvironmentVariable("OLLAMA_ORIGINS", "chrome-extension://*", "Machine")

Then fully restart Ollama (tray → Quit → start again), or reboot the machine.

⚠ Security note: chrome-extension://* allows any browser extension on the machine to call your local Ollama. For a tighter setup, replace the wildcard with ExecPilot's exact extension ID, e.g. chrome-extension://abcdefghijklmnop… (find it at chrome://extensions).

Verify it worked: open ExecPilot → Providers → Ollama should list your local models when you run any recipe. If you still see a CORS or connection error, confirm Ollama was fully restarted after setting the variable.

Simple pricing. No AI markup.

You pay your model provider directly. ExecPilot charges only for the product layer.

Free

$0

For trying ExecPilot with your own API keys.

  • 5 built-in recipes
  • Bring your own API keys
  • Chrome extension
  • Local Ollama support
Get started free

Frequently asked questions

Do I need my own API keys? +
Yes — ExecPilot is BYOM (bring your own model). You connect Claude, OpenAI, OpenRouter, ZenMux, or a local Ollama instance with your own credentials. This means no AI usage markup and full control over cost and data flow.
What is MCP and what do I need to use it? +
The Model Context Protocol is an open standard for connecting AI to tools. If your company exposes an MCP server over HTTP (e.g. a reporting or data service), add its URL under MCP Servers, approve the permission prompt, and the AI can call its tools when you ask questions. Local stdio-based MCP servers are supported in the Windows desktop app.
Can I use this without sending data to the cloud? +
Yes. Connect Ollama running locally and route any recipe to the Local tier. Files are parsed on-device before anything is sent anywhere, and local-tier recipes never leave your machine. Tool-calling also works with local models that support it (Llama 3.1+, Qwen2.5, Mistral Nemo).
What file types can I upload? +
XLSX, CSV, TXT, and MD today — with PDF, DOCX and images on the roadmap. Spreadsheets can be transformed and downloaded back as a real .xlsx file using code-execution mode, and MCP report results export to Excel too.
Can I build my own recipes? +
Yes — the Recipe Builder lets you define inputs and prompts without writing code. Pro adds output contracts, representative test cases, expected-term checks, automatic version history, and version restoration. Reusable Skills remain available for writing voice, formats, rubrics, and data-cleaning rules.
How does ExecPilot keep automated workflows under control? +
Recipe chains can pause between steps for approval. Consequential MCP tools—including create, update, delete, send, publish, approve, or payment-style actions—show the server, tool, and arguments and require explicit confirmation before execution.
Can I verify AI answers and spreadsheet changes? +
Yes. Evidence mode adds clickable source references to supported text outputs. Spreadsheet code mode provides a before/after review with changed-cell and row counts; download remains locked until the review is approved.
What can I export? +
Tabular results can be downloaded as XLSX. Pro can also export the current result or a saved decision as a branded, print-ready HTML board memo, decision paper, or presentation outline.
Ollama says connection refused / CORS error — what do I do? +
Ollama needs the OLLAMA_ORIGINS environment variable set so it accepts requests from browser extensions. See the "Ollama Setup" section above for per-session, per-user, and machine-wide PowerShell commands. Remember to fully restart Ollama afterwards.

Stop prompting. Start deciding.

Install ExecPilot, connect a model, and run your first recipe in under two minutes.

Add to Chrome — Free

How can we help?

For product assistance, licensing questions, or general support, reach out to:

help@getexecpilot.com or hello@krri.sh