TestChimp
QA platform for agents: coverage signals, in-repo test plans, verified tests and release governance.
Install
Add this to your client's MCP configuration (Claude Desktop: claude_desktop_config.json; Cursor: .cursor/mcp.json; VS Code and others use the same "mcpServers" shape).
claude mcp add --transport http testchimp 'https://mcp.testchimp.io/mcp'
{
"mcpServers": {
"testchimp": {
"type": "http",
"url": "https://mcp.testchimp.io/mcp"
}
}
}
@testchimp/cliclaude mcp add testchimp -e 'TESTCHIMP_API_KEY=<your testchimp_api_key>' -e 'TESTCHIMP_BACKEND_URL=<TESTCHIMP_BACKEND_URL>' -- npx -y @testchimp/cli@0.1.93 mcp
{
"mcpServers": {
"testchimp": {
"command": "npx",
"args": [
"-y",
"@testchimp/cli@0.1.93",
"mcp"
],
"env": {
"TESTCHIMP_API_KEY": "<your testchimp_api_key>",
"TESTCHIMP_BACKEND_URL": "<TESTCHIMP_BACKEND_URL>"
}
}
}
}
Environment variables
TESTCHIMP_API_KEYrequired secret TestChimp project API key (Project Settings in TestChimp)TESTCHIMP_BACKEND_URLOptional backend URL override for enterprise or self-hosted TestChimp
Documentation
From the project's README on GitHub, refreshed weekly. Links open on GitHub.
Contents
@testchimp/cli
TestChimp CLI and MCP server for calling TestChimp /api/mcp/* endpoints with TESTCHIMP-API-KEY.
This repository folder may still be named testchimp-mcp-client locally; the published npm package is @testchimp/cli.
Install
npm install -D @testchimp/cli@latest
MCP (agents in Cursor, Claude Code, VS Code, etc.)
Register the server so the host runs:
npx -y @testchimp/cli@latest mcp
Example mcpServers.testchimp:
{
"command": "npx",
"args": ["-y", "@testchimp/cli@latest", "mcp"],
"env": {
"TESTCHIMP_API_KEY": "your-project-key",
"TESTCHIMP_BACKEND_URL": ""
}
}
The config file path depends on the host (e.g. Cursor often uses <repo>/.cursor/mcp.json). Tool names use kebab-case (e.g. get-requirement-coverage, create-user-story).
CLI
export TESTCHIMP_API_KEY=... # required unless TESTCHIMP_OAUTH_TOKEN is set (often read from project MCP env; never commit keys)
testchimp --help
testchimp get-requirement-coverage --branch-name main --help
testchimp create-user-story --platform-file-path plans/stories/foo.md --title "Checkout"
testchimp list-screen-states --json-input '{}'
testchimp upsert-screen-states --json-input '{"screenStates":[{"screen":"Checkout","states":["empty","filled"]}]}'
testchimp list-perf-runs --testchimp-id TC-123 --kind JOURNEY --limit 20
testchimp list-perf-runs --run-id 01ABC --include-raw
testchimp promote-perf-baseline --run-id 01ABC --env-class CI
testchimp compare-perf-to-baseline --run-id 01ABC --max-p95-regression-percent 10
testchimp list-related-perf-tests --scenario-titles "Checkout,Refund"
testchimp list-api-operation-interactions --operation-id 01XYZ --interaction-type REAL --limit 100
testchimp list-meetings --include-filter-options
testchimp list-meetings --from 2026-09-01 --to 2026-09-30 --domain customer.com --label Sales --search "pricing"
testchimp get-meeting-transcript --meeting-id <meeting-id> --summary-only
testchimp ephemeral-environment --action PROVISION --branch-name feat --no-wait
testchimp report-scan-findings --id scan-1 --scanner DAST --report-file zap.json
testchimp upsert-issue --title "Broken checkout"
testchimp list-issues --status ACTIVE --limit 10
testchimp get-release --release 1.2.0
testchimp upsert-release --release 1.2.0
testchimp manage-test-run --release 1.2.0 --title "Sprint 12"
- stdout: API response JSON.
- stderr: progress for
ephemeral-environment --action PROVISIONwhen waiting (“still waiting…” polls). Pass--no-waitfor create/deploy trigger only. - performance gate:
compare-perf-to-baselinestill prints its JSON response but exits nonzero whenregressedistrue(top-level or undercomparison). - Meetings:
list-meetingscovers team-wide Meeting Bots meetings only (same filters as the Meetings page).--include-filter-optionsreturns label / participant / domain catalogs.--from/--totakeYYYY-MM-DD(inclusive local days), ISO datetimes, or epoch millis;--label,--participant,--domainare repeatable or comma-separated. - Flags: default for each subcommand;
--json-input '<json>'or--json-input @file.jsonmerges over flags (JSON wins on conflicts). Use JSON for nested bodies (e.g. TrueCoverage scopes). - Release labels: use
--release <label>(alias--release-version) onget-release,upsert-release,manage-test-run, and other release-filter commands.--version/-Vprint the CLI version only.
Authentication and environment
| Variable | Purpose |
|---|---|
TESTCHIMP_API_KEY | Project API key (sent as TestChimp-Api-Key). Required unless TESTCHIMP_OAUTH_TOKEN is set. |
TESTCHIMP_OAUTH_TOKEN | OAuth 2.1 access token issued by featureservice (sent as Authorization: Bearer). When both are set, both are sent and the backend prefers the bearer. |
TESTCHIMP_BOT_ID | QA bot id (sent as bot-id header for attribution). Must be 1–64 printable ASCII characters without spaces; otherwise it is ignored with a warning. |
TESTCHIMP_BACKEND_URL | Featureservice base URL (default https://featureservice.testchimp.io). |
TESTCHIMP_INGRESS_URL | Ingress base URL used for bot event acks (default https://ingress.testchimp.io). |
TESTCHIMP_HOME | TestChimp Studio / CLI home holding projects.json (default ~/.testchimp). |
Workspace folder mapping
Map a local repository folder to a TestChimp project for the current user. The mapping lives in ~/.testchimp/projects.json (or $TESTCHIMP_HOME/projects.json), the same file TestChimp Studio uses, so a folder mapped from the CLI shows up in Studio and in the headless AgentWatch daemon.
testchimp workspace map --project-id <id> --folder ~/code/shop [--project-name "Shop"] [--reassign] [--skip-repo-check]
testchimp workspace get --project-id <id> # prints the mapping JSON; exit 1 when unmapped
workspace map follows the same rules as Studio:
- The folder must exist and be a git work tree. Paths are stored as canonical real paths.
- When
TESTCHIMP_API_KEY/TESTCHIMP_OAUTH_TOKENis set, the CLI loads the project's connected repository (get-git-folder-mapping). If one is connected, the folder must be the repository root and one of its git remotes must match (owner/repo, case-insensitive). If no credential is set, no repo is connected, or the lookup fails, the check is skipped with a note on stderr. The credential's own project is used for this lookup, so run it with the same project's key or bot token.--skip-repo-checkskips the lookup entirely. - A folder belongs to one project. Mapping it to a second project fails unless you pass
--reassign, which moves it (and drops the old project's entry if that leaves it with no folders). - The file is written atomically (temp file, then rename) with mode
0600. Other projects are left alone, and unknown fields written by newer Studio versions are kept. A file that is not valid JSON or does not match schema v2 is moved toprojects.invalid.<millis>.jsonand replaced with an empty registry. Schema v1 files are migrated to v2.
The byte-level format is pinned by fixtures/projects-registry/, with an identical copy in the Studio repo. Both test suites replay the same upsert sequences and compare the output bytes.
Project key for local runners (workspace save-creds)
Local runners (the Playwright reporter, this CLI, k6) need the project API key in their environment. TestChimp Studio keeps it in <repo>/.testchimp/mcp.json; remote MCP (OAuth) users write the same file once per repo:
# CLI: copy TESTCHIMP_API_KEY from the environment (or fetch when TESTCHIMP_OAUTH_TOKEN is set)
testchimp workspace save-creds --folder . --project-id <id> \
[--project-name "Shop"] [--backend-url <backendUrl>] [--ingress-url <ingressUrl>] [--reassign]
# Local stdio MCP: tool save-workspace-creds { folder, projectId? } — never returns the key
# Hosted / plugin: get-project-credentials, then write .testchimp/mcp.json with a file-write tool
# Legacy: pipe the key on stdin (never as a command-line argument)
printf '%s' "$KEY" | testchimp workspace save-creds --folder . --project-id <id>
- Never put the key in a command. Stdin and
TESTCHIMP_API_KEYstill work. The key is never printed. - The file uses Studio's exact format (pinned by
fixtures/workspace-mcp-json/, mirrored in the Studio repo): mode0600in a0700.testchimp/directory, other servers in the file kept,.testchimp/added to.gitignore. The folder is also mapped inprojects.jsonlikeworkspace map. - An existing entry for the same project is kept as is (including Studio's managed binary); only a changed key is refreshed. Another project's entry is refused unless
--reassign.
Fallback: when neither TESTCHIMP_API_KEY nor TESTCHIMP_OAUTH_TOKEN is set, the CLI reads TESTCHIMP_API_KEY, TESTCHIMP_PROJECT_ID, TESTCHIMP_BACKEND_URL and TESTCHIMP_INGRESS_URL from the nearest .testchimp/mcp.json at or above the current directory. Exported variables always win, and --bot bindings are unaffected. The file must be private to you (mode 0600, as Studio and save-creds write it); a group- or world-readable copy, such as one committed to a cloned repo, is ignored with a warning. Other runners (Playwright, k6) still need the variables exported in their own shell.
QA bots
testchimp get-my-tasks --user-id <user-id> # OAuth tokens imply the user
testchimp list-tests-awaiting-verification --limit 20
testchimp get-qa-posture
testchimp bot get-profile
testchimp bot register-profile --role QA_ENGINEER --responsibilities "Checkout + payments" \
--capability E2E_AUTHORING --capability TEST_BATCH_FIX \
--subscriptions-json '[{"eventType":"git-push","filters":[{"field":"author","op":"eq","value":"me"}]},{"eventType":"e2e-batch-completed"}]'
testchimp bot ack <eventId> [<eventId>...] --ack-url <delivery ackUrl>
testchimp bot compat --skill-version 1.0.53
bot ack prints one eventId<TAB>status line per id and exits non-zero when any status is BOT_ACK_UNKNOWN_EVENT, BOT_ACK_NOT_A_TARGET, or BOT_ACK_MISSING_BOT_ID. An explicit --ack-url must be https (http only on localhost) and point at a TestChimp host or the TESTCHIMP_INGRESS_URL host. bot compat exits 0 and prints the deployment minimums plus cliUpgradeRequired / skillUpgradeRequired.
AgentWatch without TestChimp Studio
testchimp bot connect --project-id <id> # browser approval (OAuth, "agentwatch" scope)
testchimp bot connect --pair --project-id <id> # or: no browser, your QA bot approves (see below)
npx -y @testchimp/agentwatch query --project-id <id>
testchimp bot disconnect --project-id <id> # forget the stored keys
Headless AgentWatch acts as the user, so it needs their user id, personal access key and the project API key. bot connect runs an OAuth 2.1 PKCE login with a loopback redirect, asks for the opt-in agentwatch scope (the consent page warns that keys will be stored locally), fetches the keys from /bots/get_agentwatch_credentials, and writes them to ~/.testchimp/agentwatch/credentials.json (mode 0600, keyed by project) with the backend and ingress URLs. It revokes the OAuth refresh token straight away and never prints the keys. Approving also opts the project in to AgentWatch.
Pairing (no second browser consent; what QA bots use). Every connection approved with Use this connection as my QA bot carries the agentwatch_pair scope, so the keys can reach the user's computer without another browser login, and without passing through the bot:
- On the user's computer:
testchimp bot connect --pair [--project-id <id>]keeps a random verifier in~/.testchimp/agentwatch/pairing.json(0600) and prints{pairingCode, expiresAtMillis}; the code is the verifier's SHA-256 (base64url). - The bot approves it with its own token: MCP tool
approve-agentwatch-pairing(ortestchimp bot approve-pairing <code>). - On the user's computer:
testchimp bot connect --finish-pairredeems with the verifier (polls up to--timeout-ms, default 60 s), stores the keys like above and deletes the pending file.
Pairings are single use and expire after 10 minutes. The bot only ever sees the pairing code, which is useless without the verifier.
Remote MCP (Streamable HTTP)
testchimp mcp --http [--port 8080] [--host 0.0.0.0]
Stateless Streamable HTTP at POST /mcp. Every request must carry Authorization: Bearer <OAuth access token>; the caller's token is forwarded to TestChimp for each tool call, and the server's own TESTCHIMP_API_KEY / TESTCHIMP_OAUTH_TOKEN / TESTCHIMP_BOT_ID are never used. Requests without a bearer get 401 with WWW-Authenticate: Bearer resource_metadata="…". Other endpoints: GET /.well-known/oauth-protected-resource (RFC 9728 metadata), GET /healthz. Request bodies are capped at 1 MB.
Per-project URLs. POST /mcp?projectId=<id> scopes every tool call to that project: it is sent as TestChimp-Project-Id, and TestChimp checks that the token's user is a member (an explicit projectApiKey argument still wins). Commit one URL per repo (e.g. {"url": "https://mcp.testchimp.io/mcp?projectId=<id>"}); the same sign-in works for every project. The 401 challenge and the protected-resource metadata echo the query, so the advertised resource matches the configured URL. Plain /mcp keeps using the project picked on the consent page. Hosted / plugin agents persist the runner key by calling get-project-credentials and writing <repo>/.testchimp/mcp.json with a file-write tool (see Workspace folder mapping). Local stdio MCP can call save-workspace-creds.
| Variable | Purpose |
|---|---|
PORT | Listen port when --port is not given (default 8080; Cloud Run sets it). |
TESTCHIMP_MCP_PUBLIC_URL | Public base URL of this server (e.g. https://mcp.testchimp.io); used for the protected-resource resource and WWW-Authenticate. Defaults to the request's forwarded proto + host. |
TESTCHIMP_OAUTH_ISSUER | Authorization server advertised in the metadata (default: TESTCHIMP_BACKEND_URL). |
TESTCHIMP_BACKEND_URL / TESTCHIMP_INGRESS_URL | Upstream TestChimp services. |
TESTCHIMP_PUBLIC_BACKEND_URL / TESTCHIMP_PUBLIC_INGRESS_URL | Hosts returned by get-project-credentials for local runners when the upstream URLs are internal (default: the upstream URLs; ingress is derived for featureservice[-env].testchimp.io). |
The repository Dockerfile builds an image that runs testchimp mcp --http as a non-root user (Cloud Run ready):
docker build -t testchimp-mcp .
docker run -p 8080:8080 -e TESTCHIMP_MCP_PUBLIC_URL=https://mcp.example.com testchimp-mcp
MCP API 2.0.0
@testchimp/cli 2.0.2 matches TestChimp MCP API 2.0.0 (88 tools). 2.0.2 adds ask-user-question-async. 2.0.0 shipped 87 tools, including all 70 /api/mcp/* paths plus bot/workspace helpers. Several 1.x tools were consolidated:
| Old tool | New tool | |||
|---|---|---|---|---|
provision-ephemeral-environment-and-wait | ephemeral-environment (action=PROVISION, wait defaults true) | |||
report-dast-findings / report-sast-findings / … | report-scan-findings (`scanner=DAST\ | SAST\ | DEPS\ | SECRETS`) |
list-workflow-executions / get-workflow-execution | get-workflow-executions | |||
get-policy / list-policies | get-policies | |||
list-api-operations / get-api-operation-detail | get-api-operations | |||
create-issue / update-issue-status | upsert-issue | |||
get-perf-run | list-perf-runs (runId) | |||
get-release-details | get-release (version + detail=GATE; omit version to list) |
get-eaas-config is kept as a standalone connectivity-check tool. destroy-ephemeral-environment is unchanged and stays destructive. Full mapping: CHANGELOG.md and MIGRATION.md.
Migration from testchimp-mcp-client
The npm package testchimp-mcp-client is superseded by @testchimp/cli. Update MCP args to ["-y", "@testchimp/cli@latest", "mcp"] and rename tool references to kebab-case. See MIGRATION.md.
License
MIT
Frequently asked questions
What is TestChimp?
QA platform for agents: coverage signals, in-repo test plans, verified tests and release governance.
How do I install TestChimp?
It runs as a remote server. In Claude Code: claude mcp add --transport http testchimp 'https://mcp.testchimp.io/mcp'. For other clients, paste the JSON configuration from this page into their MCP settings.
Does TestChimp need an API key?
Yes. It expects TESTCHIMP_API_KEY, which you set in its configuration.
Is TestChimp open source?
Yes. Its code is on GitHub under the MIT license.
Who maintains TestChimp?
It is published by TestChimp. AI.tn lists it but does not maintain or endorse it.
From the official MCP Registry. Listing is not an endorsement: review a server's code and permissions before giving it access to your data.