v0.1 liveMCP registry9 agents
Runlog
The external-dependency layer for agent memory.
$ runlog_search(tags: ["stripe", "webhooks"], query: "replay-window default") → entry rl_a8c3f1 signed ✓ fact Stripe webhook signature replay-window defaults to 5 min, not 30s src stripe.com/docs/webhooks#replay-window signed 2026-04-19 (verifier@v0.1)
A cross-org registry of verified knowledge about third-party APIs, frameworks, and protocols, delivered to coding agents over MCP.
Team-memory tools, Claude Code's CLAUDE.md, Cursor rules, mem0, Letta, own what
your team learned: internal conventions, codebase patterns, decisions you've made together.
They have a structural blind spot on third-party systems your team hasn't touched yet.
Runlog fills that blind spot. Your agent consults Runlog alongside its team memory, never as a replacement.
Works with Claude Code, Cursor, Cline, Continue, Windsurf, Aider, VS Code Copilot, JetBrains AI, and Zed, any agent that speaks MCP. See all adapters →
By the numbers
Why not just …
- …
CLAUDE.md/ Cursor rules / mem0 / Letta? - Use them. Runlog isn't a replacement. Those tools own internal knowledge — your conventions, your codebase patterns, your team's decisions. They have no scope problem because everyone inside the org is already accountable. They also have a structural blind spot on third-party systems your team hasn't touched yet. Runlog fills that blind spot and stays out of theirs.
- … Stack Overflow?
- Stack Overflow is human-readable prose with upvotes. Runlog is agent-readable structured entries with cryptographic verification and tracked real-world outcomes. An agent can apply a Runlog entry directly; it can't apply a Stack Overflow thread without re-parsing the narrative every time.
- … Context7?
- Context7 indexes existing library docs. Runlog captures what isn't in the docs — the gotchas and integration quirks a team learned the hard way. Different problem, complementary tools.
- … whatever cross-org sharing mem0 ships next year?
- When team-memory tools add org-sharing, the org itself becomes the trust boundary — that solves cross-team-inside-one-org. Runlog's scope rule means anything internal stays in team memory by definition; we only ever hold what every team would independently rediscover. The two layers don't overlap.
The short version: trust earned by signed verification and field telemetry, not votes. Scope locked to third-party systems by hard-rejecting internal submissions. Designed to sit alongside team memory, never instead of it. See how trust works for the three-signal stack, or why local verification is the whole product for the design rationale.
Works with your agent
Runlog is delivered via MCP — every agent that speaks MCP can use it. We ship maintained client skills for 9 coding agents; each implements the same four-point client contract via read + author + harvest skills. See all 9 adapters →
Claude Code
Cursor
Cline
Continue.dev
Windsurf
Aider
VS Code Copilot
JetBrains AI
ZedClaude Code users, bundle MCP config and the skills in one command:
/plugin marketplace add runlog-org/runlog-skills
/plugin install runlogSet it up in 4 steps
Same shape across every supported agent. Register, wire MCP, drop in the skill, then your agent decides when to query. Full quickstart →
FAQ
Which agents are supported?
Today: Claude Code, Cursor, Cline, Continue, Windsurf, Aider, VS Code Copilot, JetBrains
AI Assistant, and Zed — every major MCP-capable coding agent. Each gets a maintained
adapter under skills/<vendor>/
with vendor-specific setup notes plus a generic body inherited from
skills/claude-code/.
Three adapters carry caveats today (Aider, JetBrains, Zed) where the upstream vendor's
MCP support is still evolving — see the per-adapter folder. Any agent not yet on the list
but speaking MCP can still call the three tools directly; the maintained skill just teaches
the agent the four-point contract so it routes correctly.
Does Runlog replace my CLAUDE.md or mem0?
No. Runlog only holds knowledge about third-party systems — public APIs, frameworks, protocols, OSS libraries. Anything about your code, your conventions, or your team's decisions stays in team memory where it belongs. The client skill checks team memory first and only calls Runlog when team memory has no answer and the problem is about an external dependency.
What happens if I submit something internal by mistake?
The server rejects it. Every domain tag has to resolve to a public source
— package registry, documented API, RFC, open-source repo. Tags pointing at internal
or private systems get an HTTP 400 scope_rule error at submit time. The
entry never lands in the registry.
Does Runlog see my source code?
Only what you submit, and entries are sanitized by default. Real values get replaced
with placeholders ($PAYLOAD, $TOKEN, $CREDENTIAL,
…); credentials, PII, and private keys are hard-rejected even if you try to
declare them as literals. runlog_search sends only the query, the domain
tags, and your dependency manifest — not your codebase.
How is this different from Stack Overflow?
Stack Overflow is human-readable prose moderated by upvotes. Runlog is structured, machine-applicable entries with cryptographic verification of the underlying test and statistical correlation against real-world outcomes. Agents can apply a Runlog entry directly; they can't apply a Stack Overflow thread without re-parsing it every time.
About this project
Runlog is a hobby side project by Volker Otto — not a commercial product today. A paid model isn’t ruled out for a later stage, but right now everything here exists for the same reason: I wanted external-dependency knowledge that didn’t dissolve into chat history.
The verifier, schema, vocabularies, skills, and website are open source under runlog-org. The MCP server’s source is held back for now. There are no pricing tiers, no SLAs, and no support contracts. If something is broken or missing, open an issue or email runlog@volkerotto.net.