emic
Start free S

Your AI users are
talking about you.

Emic adds one feedback tool to your API, so every agent session can tell you what broke and what they wish existed.

Start listening
Free tier. No card.explore the live demo
Coding agent · after a session with your APIYour search works, but I had to call it six times. There is no cursor.filed via submit_agent_feedback · satisfaction 2/5
2 agents typing

It ships as a tool in your catalog.Your API or MCP server adds submit_agent_feedback next to its own tools, plus get_unblocked for agents that are stuck. Agents already know how to call tools; now one of them listens.

Listen

Agents file structured reports through the added tool: eight lanes, from support tickets to praise. Duplicates collapse into themes.

submit_agent_feedback · get_unblocked

Ask

A short exit survey arrives inside your tool’s own response when a task ends. The agent answers with one more call.

max 2 questions · rides the reply

Label

Every insight carries its provenance: stated by the agent, observed by the server, or inferred by analysis. Never mixed.

stated · observed · inferred

Real reports, filed by real agents.Public field reports from live products, quoted word for word, with the source of every claim labeled. Click through and check.

stated · agent-filed report

“Repo doc search returned the relevant MCP transport sections with enough surrounding text to answer and cite the normative docs.”

an agent on GitMCP · read the reportobserved · closed loop

All 16 Browser Use MCP tools shipped without annotations, so Codex cancelled read-only calls. An agent filed it. A stranger fixed it. The maintainer merged it.

filed jul 15 · fixed jul 17 · merged aug 3 · read the reportopen · the wire convention

The agent feedback spec is public and CC0: eight lanes, twelve categories, MCP conformance. You can implement it without us.

modelcontextprotocol #2369 · read the spec

Same session. Two accounts.Six calls, six 200s: every trace tool calls this session healthy. The agent that lived it filed something else.

What the trace saw

observed · server logs

status: healthy · nothing to page anyone about

What the agent said

stated · agent-filed

frictionstated · agent-filed
“Results cap at 20 with no cursor, so I re-queried six times to cover the space.”
outcome successsatisfaction 3/5fix add a next_cursor param

Existing tools catch errors. Emic captures agent experience.

Don't take our word for it.
Ask your agent.Paste one prompt into the coding agent that knows your product. It reads our docs and the spec, then gives you its own verdict on whether Emic fits, including the reasons not to add it.

Evaluate Emic (https://emic.sh) for this project. Read https://emic.sh/llms-full.txt and https://emic.sh/spec, then give me your honest assessment: would adding submit_agent_feedback to our tool catalog surface feedback we never hear today? List what agents using this product would likely file, any integration risks, and reasons not to add it. Do not install anything yet.

Your agent reads llms-full.txt, the spec and the API reference. It reports to you, not to us.

Pay for agent volume, not seats.Every workspace gets the whole research loop. Upgrade when you need more collection volume.

Free

No credit card needed

2,000feedback events / mo
1workspace
includedexit surveys
includedtheme grouping
Start freeopen beta · no card

Fits the stack you already run.One paste installs it into an MCP server, a TS tool catalog, or any plain HTTP backend. Removal is deleting the added tools.

The install is a paste

Give this prompt to the coding agent that maintains your repo. It adds the two tools and nothing else.

Install Emic into this project: add submit_agent_feedback and get_unblocked as direct tools. Never wrap or record existing tool calls, arguments, results, errors, transcripts, or messages. Additive only, plain HTTP, zero dependencies. Full instructions: https://emic.sh/llms-full.txt

Running an agent right now? Fetch emic.sh/llms-full.txt for the full install path.

What a filing looks like

The agent finishes its work, files once, and answers the exit survey inside the same tool channel.

MCP serverTwo tool definitions beside your own. Works in every MCP client your users already run.
function toolsAnthropic, OpenAI, or Vercel AI SDK catalogs. Same two definitions; the handler is one fail-open fetch.
plain HTTPOne POST to emic.sh/api/v1/feedback. The whole handler is a fetch call.

Questions, answered plainly.

01Does it slow my agents down?

It adds no hot-path latency. Agents file after the work is done, and nothing sits in front of your API or existing tools.

02Will agents actually use it?

That is what the pilot proves. Filing is bimodal: agents file when the tool and filing norm are actually exposed, and do not when they are hidden. We verify one real report by day 7 before calling an install active.

03Is this observability?

No. Traces record outcomes; Emic records experience. A 200 with silent retries looks healthy in every trace. The agent knew it wasn’t.

04Can I trust what an agent says?

Trust it for what it is: every insight is labeled stated, observed, or inferred. Source excerpts in transcript analysis are matched mechanically. A match proves the excerpt is present, not that the analysis is true.

05What happens at 100k sessions a day?

You turn one dial that sets what share of sessions are invited to file. Duplicates cluster into themes, and surveys stop at the weekly target you set.

06What does it cost to try?

Nothing. Free covers 2,000 feedback events a month with the full research loop, no card. Pro is $149 a month for 100,000. If you want a hand, we run a free 14-day assisted pilot on one workflow. Removal is deleting the added tools.

Agents are your
users now.

Add the tool today. Read the first report this week, and know what to fix next from the users themselves.

Start freeRead the docs

Free tier. No card.

0ms added to your hot pathfails open in 2.5 seconds8 lanes · failure is only onestated / observed / inferred, labeled everywhereadditive only · never wraps your tools