> ## Documentation Index
> Fetch the complete documentation index at: https://mcpjam-mintlify-docs-update-pr-3812-1786323532738.mintlify.site/llms.txt
> Use this file to discover all available pages before exploring further.
# Evaluate
> Author test cases for your MCP server, run them across models, and gate pass/fail on deterministic checks plus optional LLM-as-judge.
A benchmark score doesn't tell you whether your server holds up in production. Evaluate lets you pin down the behaviors you care about — which tools fire, with what arguments, what the final answer looks like — and run them across the models your users actually use.
## Requirements
Evals are available to both signed-in users and guests. Guests authenticate automatically using their guest session, so no sign-in is required to run or generate tests.
## How it's organized
Holds your servers and your suites. Everything below lives in one project.
A group of cases plus the defaults they share: attached servers, models to run against, default checks, judge config, and argument-matching mode.
One scenario you want to verify — a prompt (or a sequence of prompts), the tools you expect to fire, an optional expected output, and the checks that decide pass/fail.
One execution of a suite. Produces **iterations** — one per case × model. Each iteration has its own transcript, tool calls, tokens, duration, and verdict.
## Authoring a case
Click **New case** in the suite header (or in the empty-state screen when no cases exist yet) to open a blank case editor. The case is a draft until you press **Save** — backing out without saving leaves your suite unchanged.
Each case has four moving parts:
* **Scenario** — short label so you can find it later ("Draw a rectangle", "Refuses unsafe delete").
* **Prompt turns** — one or more user messages. Multi-turn is supported; expected tools and checks attach per turn, so you can model "ask, get a result, follow up."
* **Expected tools** — for each turn, the tool calls the model should make. Arguments can be exact values or typed placeholders (`"string"`, `"number"`) so you're not chasing flaky literals.
* **Expected output** — optional free-text description of what a good final answer looks like. Used by the judge.
A **negative case** is just a case whose checks say "this tool should *not* fire." Meta questions ("what params does `search` take?"), conversational drift, and ambiguous prompts are the usual shape.
## Case attachments
Case attachments require a **pinned computer environment** on the suite. The attachments panel is hidden for draft cases and appears only after a case is saved.
Each saved test case has an **Attachments** section at the bottom of the case editor. Files you upload here are seeded into the iteration's fresh sandbox at `/home/user/attachments/` before the model's first turn runs — so the model finds the same files on every iteration, model, and re-run.
**Limits:** up to 20 files and 30 MB total per case.
To add files, click **Add file** and pick one or more files. To remove a file, click the **×** next to its name. Changes take effect on the next run.
When attachments are present, a note is automatically prepended to the case's first user turn listing each file's path, so the model knows where to find them without you having to mention it in the prompt.
**Fail-honest behavior:** if a file cannot be seeded — because its stored content is no longer available, the download fails, or a cap is exceeded — the iteration fails immediately rather than running silently without the file. This prevents a case that relies on a file from producing a misleading passing result.
## Recording widget interactions
If your MCP server uses MCP Apps (widgets rendered in a sandboxed iframe), you can record real in-widget interactions — clicks and typing — directly into a test case as `interact` steps.
When you arm record mode on a test case, a recorder shim is injected into the widget sandbox. Every click and text input you make in the widget is captured as a locator-stable `interact` step and appended to the case. The locator is built from stable identifiers in priority order: `data-testid` → ARIA role + accessible name → visible text → CSS path. This means the recorded step resolves to the same element when the case runs headlessly.
### Widget `ui/message` follow-ups
Some widget interactions send a `ui/message` back to the host (for example, a cart button that says "Show my cart") rather than calling a tool directly. During eval execution, these follow-up messages automatically drive a continuation model turn — the same way Playground handles them live. The resulting tool call (e.g. `view-cart`) is attributed to the same `interact` step's turn, so a **Tool was called** check on that turn sees the call.
This means an eval case can faithfully verify a full widget interaction sequence: prompt → widget renders → user clicks → model reacts → tool fires → assertion passes.
### What the trace shows
The Chat/Trace tab renders widget tool calls as the same app-attributed cards you see in Playground, so you can inspect exactly what the widget returned and what the model did with it.
Checks are what actually decide pass/fail. They're pure functions of the iteration transcript, so the verdict is the same every time you replay it — which is the property you want if you're using Evaluate as a regression gate.
Set defaults on the suite; override per case with **inherit** (use suite defaults), **replace** (use only the case's list), or **extend** (suite defaults, then the case's list).
| Check | Passes when | Scope |
| --------------------------------- | ------------------------------------------------------------------------------------------------------ | ------------ |
| **Tool was called with…** | The named tool was called with arguments matching your spec (per the suite's `argumentMatching` mode). | Turn or case |
| **Tool was called at least once** | The named tool fired at least once across the turn. | Turn or case |
| **Tool was never called** | The named tool did not fire. The core of negative cases. | Turn or case |
| **First tool called was…** | The first tool call in the turn matched the named tool. | Turn or case |
| **Response contains…** | The final assistant message contains a substring. | Turn or case |
| **Response matches regex…** | The final assistant message matches a regex. | Turn or case |
| **No tool errors** | No tool call returned an error. | Turn or case |
| **Final message non-empty** | The model produced a final assistant message. | Turn or case |
| **Widget rendered** | The named widget was rendered during the turn. | Turn or case |
| **Widget render latency under N** | The widget rendered within the specified milliseconds. | Turn or case |
| **Widget no console errors** | The widget rendered without console errors. | Turn or case |
| **Token budget under N** | Total token usage stayed below the cap. | Case only |
When a check is authored on an individual prompt turn it is evaluated against **that turn's slice** of the transcript — tool calls, assistant message, tool errors, and widget observations for that turn only. `Token budget under N` is case-only because per-turn token usage is not reliably captured; all other checks are turn-scopable.
### Harness system tools in assertion dropdowns
When a suite is configured to run under a harness host (such as Claude Code or Codex), the tool-name pickers in the case editor also offer the harness's native **system tools** — for example `bash`, `read`, `webSearch`, `WebFetch` — alongside your MCP server tools. This lets you write assertions like **Tool was never called** for `bash` or **Tool was called with…** for `read` without having to type the wire name by hand.
System tools appear only in assertion pickers. They are not available in pinned tool-call steps or widget-assertion "View (tool)" selects, because those require a tool that MCPJam can invoke directly.
If an MCP tool and a system tool share the same name, the MCP tool takes precedence and the system tool is not added to the list.
### Argument matching
Tool-call argument comparison runs in one of three modes, configured at the suite:
* **partial** *(default)* — every expected key must be present and match; extra keys in the actual call are ignored. Best for "I care about `query` and `limit`, not what else the model put in."
* **exact** — actual args must equal expected args, key-for-key.
* **ignore** — only the tool name is checked.
Type placeholders (`"string"`, `"number"`, `"boolean"`) match any value of that type, which lets you assert shape without locking in a literal.
### Validator settings
Each suite has default validator settings controlling how tool calls are matched. Override them at three levels:
| Level | Where | Persists? |
| ------------- | ---------------------------------------- | ------------ |
| Suite default | Suite settings → Default validators | Yes |
| Case override | Test case editor → Validators | Yes |
| Run override | Suite header → validators (sliders) icon | No — one run |
A run override shows an **override** badge; click **Reset** in the popover to clear it.
## LLM as judge
For cases where "did the right tools fire" isn't enough — anything graded on the quality of the final answer — the judge grades the run against your **expected output** if you set one, and against the user prompt otherwise. It's advisory: it produces a score and a rationale, but doesn't gate the run unless you ask it to.
* **On by default** at the suite level. The cost is gated by an explicit **Run judge** click on the run-detail page — it won't run for every iteration unless you turn auto-run on.
* **Calibrate per suite.** Judge scores aren't comparable across domains; a 0.7 on one suite isn't a 0.7 on another.
* When grading against the prompt rather than an expected output, scores are capped at 0.85 — you can't get a "perfect" without saying what perfect means.
Open a completed run and find the **LLM as Judge** panel (labelled **Goal judge** when shown inside the suite results split). Pick a judge model (default: `openai/gpt-5.4-mini`) and a threshold (default 0.7), then click **Run judge**. Each case gets a score, an advisory verdict, a one-line reason, and rubric hits. The judge never runs automatically unless you enable **Auto-run** in suite settings.
The same judge config and defaults are shared with **Swarm journeys**. When creating a new journey, expand **Advanced → Judge** to set the judge model and enable auto-grade for that journey. Session score badges (showing the verdict and score) appear in the Sessions list and run matrix cells whenever a session has been graded. Sessions with no transcript still expose the on-demand judge entry point so failed or empty sessions can be graded against the journey goal.
### MCP App widgets in swarm sessions
When a swarm session uses MCP App tools that return widgets, the session viewer replays those widgets as full interactive components — the same rendering you see in the Playground. Widgets are captured per turn as the run executes and are available once the run completes.
During a live run, MCP App tool calls appear as plain tool pills in the session pane. The interactive widget renders once the run finishes and the snapshots are persisted.
## Running
A run needs three things, all picked from the suite header:
1. **Servers** — one or more attached to the suite. Cases can attach their own subsets if they only need part of the surface.
2. **Models** — the multi-model picker is the whole point. Each model produces its own iteration per case, so you can see where Claude passes and ChatGPT trips.
3. **Run all** — kicks off every case × every model. **Run one** from a case row runs just that case.
If **Run all** is disabled, a connected server is missing, no models are selected, or a run for that suite is already in progress. The header pickers will tell you which condition applies.
### Pre-run credit estimate
When enabled on your account, an **ⓘ** icon appears beside run controls before you launch. Hovering it shows an estimated credit cost for the run about to start — priced against the exact models, cases, and iteration count the button will execute.
The estimate appears on:
* **Run all** in the suite header — prices every case × model × iterations × run plans.
* **Per-case Run** buttons in the cases list and case sidebar — prices that case's configured models for one quick run.
* **Run / Run compare** in the template editor — prices the currently selected models and the current (possibly unsaved) prompt.
* **Swarm journey cards** — prices the journey's next run across its live targets.
The tooltip shows a headline figure and a per-model or per-component breakdown. Copy varies by how well-grounded the estimate is:
| Copy | Meaning |
| ------------------------------------------------------ | ----------------------------------------------- |
| Estimated: \~N credits — based on your recent runs | Full coverage, grounded in your own run history |
| Estimated: \~N credits — partly based on recent runs | Full coverage, partially grounded in history |
| Rough estimate: \~N credits — improves after first run | Full coverage, heuristic only (no history yet) |
| At least \~N credits — part of this run is unpriced | Some models or run slices could not be priced |
| Estimate unavailable for these models | No pricing data available |
1 credit = \$0.01. Your free daily allowance and any BYOK keys change what is actually deducted. The estimate is informational only — it never blocks or gates a run.
### Frozen execution snapshots
The first run of a suite saves the set of MCP servers used as a frozen snapshot; reruns reuse it, so connecting new servers can't silently change what a suite runs against. Click **Update snapshot** in the suite header to re-save the current servers and start a new run.
## Computer environment
The computer environment picker is only visible when the **computers** feature flag is enabled on your account and a project is selected for the suite.
In suite settings, the **Computer environment** select lets you pin a built Docker environment so every eval iteration boots a fresh sandbox from the same image. This keeps results comparable across runs and edits — each iteration gets an identical starting state.
* Choose an environment from the list. Environments that have not finished building are labelled **(not built)**; selecting one will cause the run to fail immediately.
* Choose **None (default image)** to run without a pinned environment.
* The agent receives a `bash` tool inside the sandbox.
Once a run completes, the **Run detail** page shows an **Environment** row with the name of the environment used. If the environment has since been deleted, the row shows its raw identifier instead so the record remains durable. Hovering the label shows the image template, digest, and provider from the frozen snapshot taken at run time.
## Suite execution config
Each eval suite has a **Default Execution Config** section that controls the model, system prompt, temperature, tool approval, connection settings, capabilities, and host context used when running the suite.
* **Model and prompt** — Set the default model ID, system prompt, and temperature for all runs in the suite. When running evals against MCP Apps that render widgets, the eval harness automatically enables browser interaction (screenshot, click, type) for any model that supports both vision and tool calling — not just Claude. Models that lack vision or tool calling run without browser interaction.
* **Tool approval** — Toggle `requireToolApproval` to pause before each tool call during a run.
* **Server selection** — Servers are not configured here. They come from the suite's environment. The server picker is intentionally hidden in this editor.
* **Save / Reset** — Click **Save config** to persist changes. Click **Reset** to revert to the last saved state. Unsaved edits are preserved if the page refreshes the config from the server, but are discarded when you switch to a different suite.
Changes to the suite execution config apply to future runs only. Existing run snapshots are not affected.
**Personal computers are not available for eval suites.** The personal computer toggle and computer-backed tools (such as Bash) are hidden and blocked in the suite execution config editor. If the project's default host config has a computer attached, resetting a suite to that default automatically strips the computer and any computer-backed tool ids — so eval runs are never aborted by the backend's computer restriction.
### How suite defaults apply to test cases
The suite-level **system prompt** and **temperature** are runtime defaults: when a test case does not set its own system prompt or temperature, the suite values are used for that iteration. A per-case override always wins — the suite default only fills the gap.
If you have existing suites where test cases do not specify a system prompt, those cases will now run with the suite's system prompt applied. Cases that already set their own system prompt are unaffected.
## CI tab
The **CI** tab (when enabled for your project) shows only CI-active suites — suites created by the SDK or suites that CI has reported into at least once. Playground-only suites stay in the **Evaluate** tab. If a project has no CI evals yet, the CI tab shows a quickstart prompt.
The commit rail in the CI tab groups runs by commit SHA and only includes CI (SDK-sourced) runs. Playground runs on mixed suites are excluded from the rail so they don't appear as spurious manual-commit groups.
### Pass Rate vs Accuracy
The metric label shown in run headers, the runs table, charts, and hero stats follows the **run's source**, not the suite's creation source:
* **Pass Rate** — shown for runs submitted by the SDK (CI runs). Reflects per-case pass/fail.
* **Accuracy** — shown for runs triggered manually from the UI, via the API, or on a schedule.
On a mixed suite (one that has received both CI and playground runs), the label updates to match the most recent run. Legacy runs without a recorded source fall back to the suite's creation source.
### CI-synced test cases
Test cases written by a CI report carry a **CI** chip next to their name in the cases list. Opening such a case in the editor shows a notice: "Synced from CI — the next CI report may overwrite manual edits." Manual edits to these cases are preserved until the next CI ingest, which may overwrite them.
### Suite deletion
The delete button in the suite switcher is hidden for any suite that is CI-active — either created by the SDK or one that CI has reported into. This protects CI run history; the next CI report would recreate the suite anyway.
## Reading results
### Suite navigation
Opening the Evaluate tab takes you directly into the most recent suite's dashboard. To switch suites, use the **suite switcher** in the breadcrumb — it lists all your suites and lets you jump to one, create a new suite, or delete the current suite. If no suites exist yet, the empty-state create prompt appears instead.
### Suite view
* **Suite accuracy** — pass rate of the most recent run, with the last three runs' trend so you can see whether you're improving or regressing.
* **Run insights** — an AI-written diff against your previous completed run: which cases moved, which tools changed behavior, which models diverged. Skim this first; it usually points at the right rabbit hole.
* **Runs tab** — every run with its summary metrics. Click in for the iteration-level breakdown.
* **Cases tab** — every case with its latest verdict and a quick replay button.
* **Executions tab** — a flat, filterable list of every individual test execution across all cases, sorted most-recent-first. Each row shows the case name, result (passed/failed/pending/cancelled), and timestamp. Click any row to open it in the compare view.
When you run a suite against multiple host configurations at once, the Runs list groups them into a single collapsible **run group** row showing mean accuracy and longest duration across all hosts. Expand it to inspect each host's individual run.
The **Runs / Cases** segmented control switches between the run history list and the test-cases overview. **Performance by Model** in the run summary appears only when more than one model was used. To compare two runs, check their boxes in the Runs list and click **Compare**.
#### Cross-host matrix
When two or more host configurations are attached to a suite, a **By case / By host** toggle appears. **By host** shows a matrix — one column per host, one row per case. Each cell shows pass/fail dots, pass rate, median latency, and token usage. A host detached after runs were recorded stays visible, labelled **historical**.
### Run view
* **Per-iteration row** with case, model, pass/fail, tokens, duration, and the tool calls that actually happened.
* **Expected vs actual** tool calls side-by-side when an iteration fails. For failed iterations, a **Categorized diff** above the raw Expected/Actual grids groups discrepancies into four categories:
* **Missing** — expected tool calls that were never made
* **Extra** — actual calls that weren't expected (reported but non-fatal by default)
* **Out of order** — calls that happened in the wrong sequence (when order checking is enabled)
* **Arg mismatch** — right tool name, wrong arguments (shown side-by-side)
* **Full trace** — every turn, every tool call, every token. This is the thing you couldn't see before; spend time here. When a turn ran through a harness (e.g. Claude Code), the trace detail pane shows an **Engine** badge — "Claude Code" for harness turns, "Emulated" for turns that ran through MCPJam's built-in chat engine — so you can tell at a glance which runtime handled each turn.
* **Per-model breakdown** to compare how the same case behaves across models.
* **Suggested fixes** — after a run completes, a **Suggested fixes** panel ranks tool-quality and workflow issues by impact. Each issue has a **Copy** button that copies a ready-to-use fix prompt (tool description + input schema) for a coding agent; **Copy top N** copies the top issues combined.
* **Predicate Gate** — when `successPredicates` are configured on a case, an expandable Predicate Gate section in the iteration detail lists each predicate with a PASS/FAIL verdict, a one-line summary, and the evaluator's reason. Hidden when no predicates are configured.
#### Comparing two runs
Select any two completed runs and click **Compare**. The diff view shows per-case status changes: **Passed**, **Still failing**, **Regressed** (pass→fail), **Fixed** (fail→pass), **New**, **Removed**, and **Changed** (config differed). Summary metrics show deltas for tokens, cost, and duration.
### Case view
* **Pass rate across runs** — is this case stable, flaky, or trending down?
* **Pass rate by model** — does this case only work on one model?
* Every past iteration with its trace, so you can A/B a regression against a working run.
## Promoting sessions to cases
Any completed chat session can be turned into a test case without rewriting it from scratch. The full conversation is compiled into multi-turn prompt turns, and the dialog lets you pick a destination suite or create a new one.
**From the Playground** — Open the **Sessions** rail, hover a session, and click **Promote to test case**.
**From the Swarms tab** — Open a completed run, click a session row to select it, then click **Promote to test case** above the session viewer. The session must belong to a completed run attempt; sessions from in-progress runs show an error in the dialog instead of blocking at submit.
In both cases the dialog shows the servers recorded for that session, lets you set a case title and suite name, and (when attachment pickers are enabled) lets you configure the server attachment and client host for the new suite.
## Generating cases from your tools
The **Generate** button reads your attached servers' tool catalog and drafts realistic cases — a mix of positive ("call `search` with a query") and negative ("don't call `delete_user` on a meta-question"). Treat it as a draft: skim, edit the prompts to match how your users actually talk, tighten the checks, then save.
When the suite has a saved server group spanning two or more servers, the generator produces coverage for each server individually plus at least one cross-server case. Suites without a saved server group treat all available servers as a single pool.
## Plugin provenance on runs
When a run executed against an environment that pins a plugin, the run detail page shows a **Plugins** row directly below the environment label. Each chip displays the plugin name and a short bundle hash — the content-addressed identity of the exact bundle the run used.
The bundle hash is what makes two runs of the "same" plugin comparable: a re-import mints a new hash even when the name is unchanged, and an edit changes the hash under an unchanged name. The row is read-only provenance — it records what ran and offers no way to restore or re-pin those versions.
When a run is the "without skills" arm of an A/B compare, a **skills excluded** badge appears alongside the plugin chips. This distinguishes a deliberately skill-free run from one whose skills failed to load; the plugin's MCP servers were still connected for that run.
## OpenAI submission report
For runs that pinned a plugin, a **Submission report** button appears in the run header alongside **Export traces**. Clicking it downloads a Markdown document formatted for OpenAI's plugin directory review, which asks for five prompts where the plugin should be invoked and three where it must not.
The report is built from the run's own frozen case records — not from the live suite — so it names the exact bundle hash the run executed. It selects up to five passing positive cases and up to three passing negative cases. When there are not enough passing cases to fill the quota, failing cases are included to reach the count and listed explicitly at the top of the document under **Not ready to submit**.
The report also flags these conditions as not ready to submit:
| Condition | What it means |
| --------------------------- | -------------------------------------------------------------------------------------------------- |
| Fewer than 5 positive cases | The run did not execute enough "should invoke" cases. Add cases to the suite and re-run. |
| Fewer than 3 negative cases | The run did not execute enough "should NOT invoke" cases. Mark cases as negative tests and re-run. |
| Failing cases included | One or more selected cases failed; they are named at the top of the document. |
| No plugin pinned | The run pinned no plugin version, so the report cannot state which bundle was evaluated. |
| Skills excluded | The run is the "without skills" A/B arm. Submit the arm that includes skills. |
The button only appears on runs that pinned a plugin. Plugin-free runs produce no submission document.
## Exporting traces
Open a completed run and click the **Export traces** button (download icon in the run header) to download the run's traces as an OTLP JSON file. The export uses OpenInference semantic conventions plus `mcp.app.*` extensions, making it compatible with Arize Phoenix, Datadog, and any other OTLP-compatible observability backend.
### Export options
| Option | Description |
| ------------------------------- | ------------------------------------------------------------------------------------------------------------------------------------------------------------------- |
| **This run** | Exports only the iterations from the selected run. |
| **Whole project** | Exports all recorded sessions across every run in the project. |
| **Include content & artifacts** | Off by default. When enabled, the export includes prompts, outputs, tool arguments and results, and screenshots. Disable this when the content may contain secrets. |
The downloaded file is named `mcpjam-traces--.json`.
If the project is too large to export in a single file, use **This run** scope and export individual runs instead.
## What to author first
If you're new to the surface, the shortest useful loop is:
1. Attach the server you're shipping.
2. **Generate** a starting set of cases.
3. Delete the ones that don't match real usage; tighten checks on the rest.
4. Add the two or three models your users will hit.
5. **Run all**, open the failures, and decide whether the bug is in your server, your prompt, or the model.
That last step is the one that benefits most from spending time in the trace view.