docs: add weekly changelog entry (2026-08-31) - #8079
Conversation
Covers releases 2.2.37-2.2.45: headless MCP + skill pack setup, and this week's bug fixes, cost/usage accuracy improvements, and performance improvements.
github-actions
Bot
added
documentation
📋 PR Linter Failed❌ Invalid Title Format. Your PR title must include a ticket/issue number and may optionally include component tags (
Example: ❌ Incomplete Issues Section. You must reference at least one GitHub issue ( |
⏱️ pre-commit per-hook timingNo linted files changed — nothing to run. ⏭️ 44 skipped (no matching files changed)
|
|
🌿 Preview your docs: https://opik-preview-01a056e9-64e3-7740-85e1-c8cf5b5dc1db.docs.buildwithfern.com/docs/opik The following broken links were found: ❌ Broken link: 2.2.36...2.2.45 (404) Unverified links (timeout / rate-limited / server error — not failing the check)• https://aistudio.google.com/apikey (401) 📌 Results for commit 63e9476 |
![baz-reviewer[bot]](https://avatars.githubusercontent.com/in/933528?s=60&v=4){kind=small}
| @@ -0,0 +1,37 @@ | |||
| ## MCP + Skill Pack Setup Can Now Run Without a Terminal | |||
|
|
|||
| Setting up the Opik MCP server previously meant sitting through an interactive wizard, and it only reached three of the AI assistants people actually use. `opik configure --install-mcp` and `opik mcp configure --ai-client <host>` now run non-interactively — pass the client explicitly (or `--ai-client all`) and the command completes on its own, so it can run from a coding agent, a Dockerfile, or CI. Codex and opencode join Claude Code, Cursor, and VS Code Copilot as supported hosts, and every install now ends with a real verification call that reports the workspace and project count instead of assuming the config it just wrote actually works. | |||
There was a problem hiding this comment.
Hosted installs falsely promise project counts
The changelog claims every install reports a workspace/project count, but hosted installs use verify_hosted_endpoint only to validate reachability and the expected 401/403 auth challenge, so the count claim is misleading — should we qualify it as local-credential verification and describe hosted verification as endpoint/auth-challenge validation?
Want Baz to fix this for you? Activate Fixer
Other fix methods
Prompt for AI Agents
Before applying, verify this suggestion against the current code. In
`apps/opik-documentation/documentation/fern/docs-v2/changelog/2026-08-31.mdx` around
line 3, revise the MCP verification description to accurately distinguish the two
installation paths. State that local credential verification calls the projects endpoint
and reports the workspace/project count, while hosted verification only validates
endpoint reachability and the expected authentication challenge; do not claim that
hosted installs report project counts.
|
|
||
| Setting up the Opik MCP server previously meant sitting through an interactive wizard, and it only reached three of the AI assistants people actually use. `opik configure --install-mcp` and `opik mcp configure --ai-client <host>` now run non-interactively — pass the client explicitly (or `--ai-client all`) and the command completes on its own, so it can run from a coding agent, a Dockerfile, or CI. Codex and opencode join Claude Code, Cursor, and VS Code Copilot as supported hosts, and every install now ends with a real verification call that reports the workspace and project count instead of assuming the config it just wrote actually works. | ||
|
|
||
| A new `--install-skills` flag (and the equivalent `--skills` on `opik mcp configure`) installs the companion Opik skill pack alongside the MCP server — instructions the assistant reads to know how to use the tools the server exposes, rather than just their tool list. The two are asked about separately since they carry different trust implications: the MCP server writes credentials into a config file, the skill pack installs instruction files the assistant executes with its own permissions. The onboarding copy-paste prompt and docs now point at `/opik-instrument` rather than the unnamespaced `/instrument`, avoiding collisions with other skills of the same name. |
There was a problem hiding this comment.
Unattended skill install hides mutable code source
The trust guidance omits that setup_skills calls fetch from https://codeload.github.com/comet-ml/opik-skills/tar.gz/refs/heads/main, a mutable ref, so unattended runs in CI/Dockerfiles/agents pull whatever's on main at execution time, and write_skill silently replaces existing skill directories — should we document the repository/ref and replacement behavior?
Want Baz to fix this for you? Activate Fixer You can also update your AI coding guidelines based on this comment by apply pr to [branch name]
Other fix methods
Prompt for AI Agents
Before applying, verify this suggestion against the current code. In
`apps/opik-documentation/documentation/fern/docs-v2/changelog/2026-08-31.mdx` around
lines 5-5, update the MCP and skill-pack trust guidance to disclose that `setup_skills`
downloads instructions from the `comet-ml/opik-skills` repository at the mutable
`refs/heads/main` ref. Explicitly state that installation removes and replaces existing
skill directories before linking the downloaded skills into assistant locations,
especially since the commands are recommended for unattended use in coding agents,
Dockerfiles, and CI.
|
|
||
| Setting up the Opik MCP server previously meant sitting through an interactive wizard, and it only reached three of the AI assistants people actually use. `opik configure --install-mcp` and `opik mcp configure --ai-client <host>` now run non-interactively — pass the client explicitly (or `--ai-client all`) and the command completes on its own, so it can run from a coding agent, a Dockerfile, or CI. Codex and opencode join Claude Code, Cursor, and VS Code Copilot as supported hosts, and every install now ends with a real verification call that reports the workspace and project count instead of assuming the config it just wrote actually works. | ||
|
|
||
| A new `--install-skills` flag (and the equivalent `--skills` on `opik mcp configure`) installs the companion Opik skill pack alongside the MCP server — instructions the assistant reads to know how to use the tools the server exposes, rather than just their tool list. The two are asked about separately since they carry different trust implications: the MCP server writes credentials into a config file, the skill pack installs instruction files the assistant executes with its own permissions. The onboarding copy-paste prompt and docs now point at `/opik-instrument` rather than the unnamespaced `/instrument`, avoiding collisions with other skills of the same name. |
There was a problem hiding this comment.
Hosted MCP trust model is misstated
The doc claims the MCP server writes credentials into a config file, but detect_hosted_mcp_server only builds URL-only RemoteServerSpec blocks — credentials are stored only for local stdio registration — so readers may not realize hosted registration (the default for opik mcp configure) instead relies on the host's browser OAuth flow, should we qualify the trust note by transport?
Want Baz to fix this for you? Activate Fixer You can also update your AI coding guidelines based on this comment by apply pr to [branch name]
Other fix methods
Prompt for AI Agents
Before applying, verify this suggestion against the current code. In
apps/opik-documentation/documentation/fern/docs-v2/changelog/2026-08-31.mdx around line
5, revise the MCP-versus-skill-pack trust explanation to distinguish hosted and local
transports. Clarify that hosted MCP registration records the server URL and relies on
the host's browser OAuth flow, while local stdio registration stores API credentials and
workspace values in the configuration; retain the warning that skill-pack instructions
run with the assistant's permissions.
|
|
||
| - **Faster dataset experiment summaries** — The query behind a dataset's experiment summary scanned every experiment item in the whole workspace before filtering down to the requested dataset. It now prunes to the relevant experiments upfront, cutting rows read by roughly 13x on a workspace with 3M experiment items across 200 experiments, and skipping the scan entirely for a dataset with no experiments. | ||
|
|
||
| - **Faster dataset item inserts** — Adding items to a dataset updated the version's item count with a read-modify-write cycle inside the per-dataset lock. It's now a single atomic increment, cutting the database round trips on that path from three to one. |
There was a problem hiding this comment.
Overstated dataset insert round-trip reduction
“On that path” implies the full insert request drops from three database round trips to one, but existing-version inserts still call countExistingItemIds, insertItems, and updateVersionCountsForInsert — can we scope the claim to version-counter maintenance and say, “the version item-count update now uses a single atomic increment instead of a three-round-trip read-modify-write cycle”?
Want Baz to fix this for you? Activate Fixer
Other fix methods
Prompt for AI Agents
Before applying, verify this suggestion against the current code. In
`apps/opik-documentation/documentation/fern/docs-v2/changelog/2026-08-31.mdx` around
line 31, revise the “Faster dataset item inserts” description so the three-to-one
reduction clearly refers only to version item-count maintenance, not the complete item
insertion request. State that the version item-count update now uses a single atomic
increment instead of the previous three-round-trip read-modify-write cycle, while
preserving the rest of the performance context.
2 participants
Details
Adds the weekly product changelog entry covering releases 2.2.37–2.2.45. Leads with the headless MCP + skill pack setup (
opik configure --install-mcp --install-skills,opik mcp configure --ai-client <host>), and consolidates the week's bug fixes, cost/usage accuracy fixes, and performance improvements. The "agentic tools" feature-toggle removal is filed under Bug Fixes since the feature itself was already announced in an earlier changelog entry.Excluded from this entry after verification against the code:
redaction.enabled) — explicitly beta and off by default inconfig.yml, not shipped to any users yet.cipx_savings_factstable, and the various[OPIK-7315]/cutover commits — internal migration/ops tooling, no user-visible surface.google-cloud-vertexaibump for Gemini "thinking" config — only unblocks the dependency; nothing in the codebase wires the new field up yet.Change checklist
Issues
AI-WATERMARK
AI-WATERMARK: yes
main; verified CLI flag names, feature-flag status, and technical claims directly against the current source before including them.Testing
Documentation-only change (new
.mdxfile). No code paths affected.docs.yml(changelog: ./docs-v2/changelog), so no additional registration was needed.sdks/python/src/opik/cli/{configure,mcp}.py, theredaction.enableddefault inapps/opik-backend/config.yml, the Bedrock cache-token diff, etc.) rather than relying on commit messages alone.Documentation
This PR is the documentation change — new entry at
apps/opik-documentation/documentation/fern/docs-v2/changelog/2026-08-31.mdx.Generated by Claude Code