INFERENCEMCPBETA

Inference MCP plugs your workspace into any LLM client. Ask about projects, inferences, traces, evals, and training jobs in plain language, from the coding agent you already use. OAuth sign-in in the browser. No API key.

https://mcp.inference.net/mcp
Install in Cursor
> why did checkout latency spike at 3am?
✳ list_traces p95 > 2s, 03:00-04:00
└ 3 slow spans, all in gateway/retry. Provider timeout,
failover added 1.9s. Full trace: tr_8f2a...

Debug production traffic

Ask why latency spiked. Get the exact trace back, not a dashboard safari.

food-extraction87.0
schema-parsing ▼ 12 samples regressed84.1
tool-call-accuracy91.4

Interrogate your evals

Scores dropped overnight? Ask which samples regressed and read the diffs without leaving your editor.

schematron-v3-smalltraining
step 4,200 / 6,000loss 0.213

Watch training jobs

Poll your fine-tune from the same session that launched it. Loss curves as text beat tab switching.

> how much did glm-5.2 cost us last week?
✳ team_get_usage_stats model=glm-5.2, range=7d
└ $1,284.10 across 3.2M requests. Cache hits saved
$310. Cheapest day: Sunday. Want the daily split?

Pull the numbers

Spend, tokens, cache hits, latency percentiles. Ask for the figure instead of exporting a CSV.

How it works

01

Add the server

One command or one click. Works in any MCP client that speaks HTTP.

02

Sign in once

OAuth in the browser. Your session gets your team permissions, nothing more. Revoke anytime under Settings → Connected Agents.

03

Ask

Your agent picks the right tool per question. You read answers, not dashboards.

The toolbox

Tool names may change while in beta.
list_projects

Every project you can see, with traffic summaries.

list_inferences

Filter request logs by model, status, latency, or time.

get_trace

Full span tree for one request, timings included.

list_eval_runs

Eval runs with scores, deltas, and regressed samples.

get_training_job

Status, steps, loss, and artifacts for a fine-tune.

list_halo_agents

Your Halo agents and their recent runs.

list_signals

Alerts and anomalies your project has raised.

team_get_usage_stats

Spend and token counts by model, project, or day.

list_datasets

Datasets built from captured traffic, ready to train on.

Questions

Is this read-only?

Reads by default. Write scopes are requested at sign-in and follow your team role. A viewer stays a viewer.

Does it work with my client?

Any MCP client that speaks streamable HTTP. The six above are tested. Everything else takes the raw URL.

What does it cost?

Nothing. The MCP server is free. You pay only for inference you run on the platform.

How do I revoke access?

Settings → Connected Agents lists every connected client. Revoke one and it is signed out on its next call.

Your workspace, one prompt away.

https://mcp.inference.net/mcp
Install in Cursor

Two minutes. OAuth only. Uninstalls as fast as it installs.