Quickstart Get API key

Agents that work with Runlog

Runlog is delivered via MCP — every agent that speaks MCP can use it. We ship maintained client skills for 9 coding agents. Each adapter pairs the same four-point client contract (check team memory first, classify external-deps, route learnings correctly, track the dependency manifest) with vendor-specific MCP setup and rule-loading glue. Three skill types ship per vendor: a read skill that consults the registry, an author skill that submits a verified entry mid-flow right after a third-party gotcha is solved, and a harvest skill that scans the just-finished session at the end and routes any missed external-dependency findings through the same submission flow.

Both forms are ephemeral — nothing installed globally:

npx (Node)

Claude Code, Cursor, Cline, Zed

npx add-mcp https://api.runlog.org/mcp

Pass -a claude-code, -a cursor, -a cline, or -a zed to target one host; omit to install across every detected agent.

Note: add-mcp uses claude-code as the host key; the runlog-installer uses claude (matches runlog install --target claude). Both target the same host with different naming conventions — pick whichever suits your workflow.

pipx run (Python)

Aider, Continue, Copilot, Windsurf

pipx run --spec runlog-installer runlog install \
  --target <host> --api-key $RUNLOG_API_KEY

The --spec flag tells pipx the package name; the runlog after it is the script.

Prefer a persistent install? pipx install runlog-installer then runlog install --target <host> --api-key $RUNLOG_API_KEY — same flags, leaves the binary on $PATH.

JetBrains AI Assistant: manual config — see the per-vendor Quickstart below (config format is XML / IDE-specific; not yet supported by the installer).

Claude Code users can take an even shorter path that bundles MCP config and skill bodies in one command:

/plugin marketplace add runlog-org/runlog-skills
/plugin install runlog

Runlog is listed in the Official MCP Registry: npx add-mcp find runlog resolves the entry at registry.modelcontextprotocol.io and writes the right config for your detected agents — an alternative to the direct-URL form above when you'd rather discover by name.

Claude Code

Reference adapter.

Read + author + harvest skills. Teaches the four-point contract (check team memory first; classify external-deps; route learnings; track + report).

~/.claude/skills/runlog/

Cursor

Highest priority after Claude Code.

Read + author + harvest skills. Teaches the four-point contract (check team memory first; classify external-deps; route learnings; track + report).

.cursor/rules/runlog.mdc

Cline

Open-source, MCP-native.

Read + author + harvest skills. Teaches the four-point contract (check team memory first; classify external-deps; route learnings; track + report).

.clinerules/runlog.md

Continue.dev

Open-source, MCP-native.

Read + author + harvest skills. Teaches the four-point contract (check team memory first; classify external-deps; route learnings; track + report).

.continue/config.yaml

Windsurf

Codeium-based, Cascade agent.

Read + author + harvest skills. Teaches the four-point contract (check team memory first; classify external-deps; route learnings; track + report).

.windsurfrules

Aider

CLI-native, diff-cycle model.

Read + author + harvest skills. Teaches the four-point contract (check team memory first; classify external-deps; route learnings; track + report).

CONVENTIONS.md or --read

* MCP support is version-dependent

VS Code Copilot

Requires Copilot agent mode.

Read + author + harvest skills. Teaches the four-point contract (check team memory first; classify external-deps; route learnings; track + report).

.github/copilot-instructions.md

JetBrains AI

IntelliJ, PyCharm, WebStorm, GoLand, …

Read + author + harvest skills. Teaches the four-point contract (check team memory first; classify external-deps; route learnings; track + report).

AI Assistant guidelines

* Tool-use varies by IDE / plugin version

Zed

Native context_servers + .rules.

Read + author + harvest skills. Teaches the four-point contract (check team memory first; classify external-deps; route learnings; track + report).

~/.config/zed/settings.json

* HTTP context_servers schema is settling

Asterisks flag adapters whose upstream vendor's MCP integration is still evolving. The skill body works against today's vendor capabilities; the per-adapter folder has VERIFY notes flagging where to re-check the vendor's current docs before publishing your config.

Cross-vendor invariants — the four-point client contract plus the author and harvest contracts every adapter inherits — live in skills/common/. When a contract rule changes, one file moves and every adapter re-syncs.

At the end of a debugging session that touched a third-party system, your agent surfaces a one-line nudge — Harvest 2 candidate findings? (y/N) — and on yes routes each candidate through local Ed25519 verification before drafting a runlog_submit call.

/runlog:harvest
Scanned this session + last 3 commits. Candidates:
  1. Stripe webhook signature: replay-window default is 5 min, not 30s [score 0.86]
  2. Notion API: Retry-After header on 429 responses is bounded; bursts beyond ~1000 req/hr block silently [score 0.74]
Pick (e.g. 1,2 or all): _

Each pick re-runs locally against the verifier, signs the result, and submits a draft entry — the registry never sees an unverified claim.