YBacked by Y Combinator


for agent sessions.

When an agent uses your product, the session lives inside Claude, not your UI. Armature captures it before it vanishes.

About us
See sessions from every agent client
Claude Code Codex OpenClaw Claude ChatGPT Gemini CLI OpenCode Cursor
The problem

You see the calls. Not the agent.

What you can't see Their AI client
90%
prompt
reasoning
satisfaction
↺ Retried 4 times. Gave up.
Replied "I can't help with that."
↗ Suggested switching to a competitor's MCP.
What you can see Your MCP server logs
10%
14:32:08 → mcp_server.connect_db({db: "users-prod"}) 200 OK

This is happening right now to your users, and your logs will never show it. Armature captures the whole session, the user's intent, the agent's thinking, every tool search, and brings visibility back.

Features

Everything you need for User Agent Experience.

01 · Analytics

Analytics for every agent session.

Reconstruct what agents really do with your product: the intent, the reasoning, every tool call, success or failure. Turned into funnels and dashboards.

02 · Interface

We generate code-mode MCPs.

From your API, we build a top-performing code-mode MCP, host it, and keep it in sync with every release.

03 · Testing

Tests and benchmarks on every harness.

Run complex workflows across every model and harness, and benchmark your agent experience against competitors.

04 · And more

Every growth lever, rebuilt for agents.

The loop doesn't stop at analytics. Alerts, auto-fixes, experiments, lifecycle, upsell, every lever you had for humans, rebuilt for a world with no human at the screen.

Alerts
Monitoring
Suggestions
Auto-fixes
Experiments
Lifecycle
Upsell
Discovery
More soon
Get started

Live in minutes.

Recommended

Hosted code-mode MCP

We generate and host it: optimized code-mode, always in sync, and you own tool search.

plug into your CI/CD → get a URL

Wrap your own MCP

Keep your MCP. Wrap it with our SDK and start capturing sessions.

armature.wrap(yourMcp)
You’re all set. Sessions, analytics, and alerts start flowing in.
About us

We've shipped both halves. Now the missing piece.

We've deployed MCPs used by millions of customers and built observability products from the ground up. Now we're building the missing piece, to help builders ship a seamless user agent experience.

Backed by top investors including Y Combinator.

Palantir Tsuga Joko Y Combinator
More about us →
FAQ

Common questions before you start.

How is this different from PostHog, Amplitude, or Mixpanel?
Those track humans clicking a UI. When a user delegates to an agent, the session moves into Claude or ChatGPT, outside your product, so those tools see nothing. Armature is built for that world: it captures the agent session, the intent, the reasoning, the tool calls, that human-analytics tools can't reach.
How is this different from LLM observability like LangSmith or Langfuse?
Those observe the agents you build, for engineers optimizing your own AI features. Armature is for product and growth teams: it shows how your users' agents experience the product you ship, your MCP or CLI. Different user, different question.
Do I have to replace my MCP?
No. Two ways in: we generate and host an optimized code-mode MCP (recommended, so you own tool search and stay in sync), or you keep your own MCP and wrap it with our SDK. Either way you start capturing sessions.
What can you see that my server logs can't?
Your logs show isolated tool calls, a clean 200 OK even when the user is giving up. Armature reconstructs the whole session: the user's intent, the agent's reasoning, the tool searches, and whether they actually got what they came for.
Do you still do testing and benchmarks?
Yes. We run synthetic sessions across every model and harness to catch broken flows before users do, and benchmark your agent experience against competitors. It's part of the platform, analytics is the core.
Which agents and clients do you support?
Any client your users bring, Claude, ChatGPT, Cursor, Codex, Gemini, and more, and any MCP (HTTP, SSE, stdio) or CLI. Bring your own auth; secrets stay server-side, never exposed.

See how agents really use your product.

Generate a hosted code-mode MCP, or wrap your own with our SDK. Start capturing every agent session today.