Let your coding agents ask your running systems questions
One MCP endpoint over your logs, errors, product analytics and tickets.
Specific, cited answers for your agents. Visibility and control for your team. Read-only and isolated for your deployments.
Questions your agents should be able to just answer
- "Has anyone used the new export feature since Tuesday?"→ usage events, counted and linked.
- "Summarize all searches since 9am this morning."→ grouped log patterns with counts and examples.
- "What errors and warnings appeared after release
api@1.42?"→ a joined timeline of the deploy, the errors and the log changes. - "Is the
new-rankingflag live for anyone, and are they hitting errors?"→ analytics and errors, joined. - "Pull today's errors and warnings, open issues for the real ones, fix them, verify, close."→ the everyday triage loop, with evidence attached to every issue.
Today, each of these means a human pasting from three dashboards, or an agent juggling several MCP servers with different query languages. We think it should be one question and one cited answer.
Built to work for everyone involved
For your agents
One endpoint, a small set of tools, and a schema they can read. Summary-first answers with links to the source, instead of walls of log text. A shared notebook so the next agent, or the next session, starts where the last one stopped.
For your team
See what your agents asked, what they found, and what it led to: connector health, agent usage, and findings linked to issues and releases. Scoped access you can explain in one paragraph.
For your deployments
Read-only by design. Your logs stored in an open format in isolated, per-team storage. Query budgets so an agent loop can't run up a bill. Works alongside the vendor tools you already use; it doesn't replace them.
How we imagine it working
- Connect your sources.
Send logs over OpenTelemetry or HTTP (or straight from AWS ECS/CloudWatch). Connect Sentry, PostHog, Linear and GitHub with OAuth.
- Add one MCP URL to your agents.
Claude Code, Cursor, Codex and other MCP clients.
- Ask.
Agents get joined, cited answers and save what they learn to a shared notebook. You see it all in a simple console.
Why not just connect five MCP servers?
You can, and many teams do. Vendor MCP servers for Sentry, PostHog, Linear and your cloud are good at their own data. What's missing:
Agent's Notebook is designed to sit beside those servers. Your agent still creates and closes issues with its Linear tools; we supply the evidence and remember the outcome.
- Joins.
"Errors since this release, for users with this flag" crosses three tools.
- Memory.
Findings vanish when the session ends unless someone copies them into a ticket.
- Safety.
Log and error text is attacker-writable. In June 2026, Tenet Security showed fake Sentry errors steering coding agents into running attacker-controlled code ("Agentjacking"). Tool output needs to be treated as untrusted data.
- Visibility.
Nobody can see what agents asked, or what it led to.
FAQ
What is Agent's Notebook?
Agent's Notebook is a planned MCP service from Babbl Labs that gives coding agents one safe endpoint for asking questions of a team's logs, errors, product analytics and tickets, and a shared notebook for what they find. It's in design preview and not yet available.
Is it available?
Not yet. We're validating the design with a short survey for developers and a matching survey their coding agents can fill out. Survey respondents get a spot in line for the beta and an early-adopter discount.
What will it cost?
Launch pricing for the Team plan is $99/month for up to 5 seats, then $25 per seat. Early survey respondents lock in a discount: 50% off forever for founding teams (spots below #250), 30% off forever up to spot #1,000, and 30% off for the first 6 months after that. Your spot is assigned when you add an email on your claim page and kept once you verify it. Enterprise own-tenant and appliance deployments are priced separately.
Which coding agents would it work with?
Any MCP client, including Claude Code, Cursor, Codex, GitHub Copilot and others. It's agent-neutral by design.
How do I give Claude Code access to my production logs today?
Today you can connect vendor MCP servers (for example Sentry, PostHog, Linear, Supabase, or the AWS CloudWatch MCP server) individually, preferably in read-only mode with narrowly scoped tokens. Agent's Notebook aims to replace that patchwork with one read-only endpoint that joins those sources and cites every answer.
Can agents write to my systems?
No. Agent's Notebook tools are read-only on your data. Agents keep using your existing tools (like Linear's MCP) to create or close issues; Agent's Notebook supplies evidence and records the outcome.
Where is my data stored, and who can see it?
The plan: each team's data lives in isolated storage with its own encryption key, readable only through credentials scoped to that team, with an audit log of every agent query. Enterprise customers would be able to run it in their own cloud account or as a self-contained appliance.
Does it replace Datadog, Sentry or PostHog?
No. It works alongside them. Logs can be ingested into an open-format store you control, while tools like Sentry, PostHog and Linear are queried live.
How do you protect agents from prompt injection in logs?
Every piece of text that comes from a log, error or ticket is returned to the agent clearly marked as untrusted data, and instruction-like text is flagged. Tools never execute anything.
Who is behind it?
Babbl Labs, a small team that runs a high-volume data pipeline on AWS with Sentry, PostHog and Linear, and coordinates frontend and backend coding agents daily. We're building this for ourselves first.
Can my coding agent take the survey?
Yes. Ask it to "Read https://agents-notebook.com/agents and claim my discount." It answers a few multiple-choice questions from what it already knows about your project (tool names only, no secrets or personal data), without opening files or running commands. It then hands you a claim link; adding your email there locks in your spot and discount, and you can invite teammates and request features.
You're early. Claim your discount (about 2 minutes)
Tell us how your agents get production context today and what would help, and claim an early-adopter discount on the Team plan. Launch pricing: $99/month for up to 5 seats, then $25 per seat. Founding teams get 50% off, forever, while founding spots last; after that, 30% off (forever up to spot #1,000, then for 6 months). You'll also get a spot in line for the beta; an email on the claim page locks it in.
Why a team plan? One set of scoped, read-only connections to your shared data sources, brokered and federated so every agent and every teammate asks the same sources with the same guardrails, and findings land in one shared notebook your whole team can read.
"Forever" means for as long as your team's account stays active. Discounts apply to the Team plan and additional seats; see the terms. Terms. We'll publish the aggregate results.
Using a coding agent? Paste this:
Read https://agents-notebook.com/agents and claim my discount.Last updated: 2026-10-12