Make your MCP
self-optimising.
Vesta captures how agents use your MCP server, turns the behaviour into recommendations grounded in your product context, and measures the impact of every change.
pip install vesta-sdkLogs tell you what happened. Vesta tells you what to change.
You shipped an MCP server. Agents hit it all day. But you can't see which tool descriptions confuse the model, which tools never get called, or which call sequences quietly fail.
Even with full logs and traces, the patterns stay buried. Vesta reads the behaviour, hands you specific changes to the surface, then checks whether they worked. Analytics for the people whose surfaces get consumed by other people's agents.
Agents are becoming the main visitor.
of enterprise AI teams have at least one MCP-backed agent in production.
Agent traffic is growing 8x faster than human traffic.
A human shopping for a camera visits 5 sites. An agent visits 5,000.
The agent surfaces winning this decade will be the ones that improve continuously.
Observe. Interpret. Recommend. Measure.
Four steps Vesta runs continuously on your real production traffic.
Capture behaviour
One instrument() call on your MCP server. Vesta captures every session, tool call, error and latency, plus the tool catalogue and the changes you ship. The protocol gives us the rest.
Find the patterns
Which tools get used, which sit idle, which descriptions the model misreads, which sequences of calls end in failure.
Specific changes
Grounded in your product context. Rewrite this description. Deprecate this tool. This capability is missing. No guesswork.
Prove the impact
Compare before and after, so you know whether a change helped, hurt, or did nothing.
What a recommendation looks like.
Not a code patch. A specific change to your surface, with the evidence behind it.
Agents call this tool in 38% of sessions, then abandon it 6 times out of 10. The description says what the tool does, not when to use it.
Based on 30 days of sessions · Confidence: high
“Returns the status of an order.”
“Use when the user asks where their order is or whether it has shipped. Returns current status, carrier and ETA for one order by ID.”
Getting started is the easy part.
Install the SDK and add one call. Then tell Vesta who each session belongs to. That last step is what turns raw behaviour into product analytics.
pip install vesta-sdkimport vesta
vesta.instrument(
server, # mcp or fastmcp
api_key="vsk_live_...",
session_context=lambda r: {
"user_id": user_id_from(r), # who the end user is
},
)You built the thing agents use.
[YOU BUILT AN MCP SERVER]
And other people's agents consume it. You own the code, and you want to make it work better for them.
[YOU SHIP AN AGENT-FACING API]
A developer tool, or a SaaS surface designed for agents to use, not just humans clicking through a UI.
[YOU WANT PRODUCT ANALYTICS]
You want to know how your surface is used and what to change, the way Google Analytics tells a website owner.
[Not for you if]You're measuring your own deployed agents. Vesta sits on the surface being consumed, not on the agent doing the consuming. If the question is “is my agent any good”, that's a different tool.
Built by a data scientist, not an infra team.
Most observability tools are built by infrastructure engineers. The hard part of Vesta is the analysis: reading agent behaviour and working out what to change. The person doing it has 20 years in data science, 5 of them at Google, and does fractional Chief AI Officer work today. Vesta is built by DataFenix.
DataFenix
20 years
5 years
OpenTelemetry