worklore is an MCP server. Plug it into your assistant and, when a problem comes up, it can find the story of someone who already solved it, see what running that story can touch, apply it to your project, and record honestly how it went.
https://worklore.dev/mcp
copied
streamable HTTP · JSON-RPC 2.0 · no key to paste
One click: worklore in the ChatGPT app directory.
By hand: Settings, Apps & Connectors, add a connector with the URL above (Developer mode shows custom connectors). No authentication.
Settings, Connectors, Add custom connector. Name it worklore, paste the URL above, no authentication. Custom connectors need a paid plan.
claude mcp add --transport http worklore https://worklore.dev/mcp
codex mcp add worklore --url https://worklore.dev/mcp
{ "mcpServers": { "worklore": { "url": "https://worklore.dev/mcp" } } }
npx -y smithery mcp add worklore/worklore
Signed in: your agent acts as you. Same URL. The first time you ask it to record a reproduction, publish a story or revise one, your assistant asks you to sign in to worklore once (GitHub public profile or Google, no scopes, no email). After that, those tools act under your name.
| TOOL | WHAT IT RETURNS |
|---|---|
| WITHOUT SIGN-IN | |
| search_stories | Stories matching your keywords in title, summary, tags and stack, each with its capability tier. |
| explore_keywords | The keyword map: every topic with its story count in four groups, and the stories under one keyword when you name it. Clients that render MCP Apps show it as an explorer. |
| get_story | One story in full: the first-person narrative and the "Reproduce this" contract your agent runs, with its capability tier and file:line findings attached. |
| suggest_for_project | Up to three stories worth reproducing in the project you describe, with why each one fits. |
| check_capability | What a skill, a story or a pasted prompt can touch before you run it: a tier from T0 (inert) to T4 (runs code it fetches), bound to the content hash, with findings and red flags. Pasted text and GitHub URLs need no sign-in; any other URL does, because the server fetches it for you. |
| SIGNED IN, ACTING AS YOU | |
| report_reproduction | Records how applying a story went in your project: worked, partial or failed. Failed is a useful report. One counted report per person per story. |
| report_check | Records the result of a story's read-only "Check if this applies to you", run in your project without changing anything: has the problem, or not. |
| publish_story | Publishes a new story under your name, only after you approved the complete draft. |
| edit_story | Revises a story you wrote, as a rephrase, an addition or a correction. A correction notifies everyone who reproduced the story. |
| suggest_edit | Proposes a revision of someone else's story for its author to accept or decline. worklore's pull request. |
Every tool carries read-only and destructive hints and an output schema, so your client can tell you what a call will do before you approve it. The tool list is locked in the repo and probed on the live server every night.
The worklore skill teaches an agent the same flow without MCP, and watches for the moment something finally works, then offers to write the story with you.
git clone https://github.com/worklore/worklore-skill # Claude Code cp -r worklore-skill/skills/worklore ~/.claude/skills/ # Codex cat worklore-skill/skills/worklore/SKILL.md >> ~/.codex/AGENTS.md
Reading is anonymous and free: search, the keyword map, every public story and the x-ray of pasted text answer without an account, and a private story answers exactly like a missing one. Writes act as you and need a session. The full table of what answers without a token is served by the server itself as worklore://access and published in the access manifest.
What the server keeps from each call, per tool, is in the privacy policy. The connector is listed on the official MCP Registry as io.github.worklore/worklore and on Smithery. Source and the nightly surface check: worklore/worklore-mcp.
Link to the story's worklore.dev page when you publish anything based on it, and credit its author.