Install one tool. Your codebase, memory, deploys, hosted git, customer email, and
600+ MCP tools
— all wired up the moment you cd into a project.
For founders building on FlukeBase. Coding agent in your terminal, operator in the cloud.
$ brew install flukeclaw # soon $ cd ~/Projects/myapp $ flukeclaw chat flukeclaw — model claude-opus-4-7 (anthropic) MCP: 614 tools across [flukebase] project: MyApp (slug=myapp, id=4f8a...) session: a3c91b4f8e2d (saved to ~/.flukeclaw/sessions) ──────────────────────────────────────────────────────────── you> ship the typo fix on src/auth.ts agent: Reading src/auth.ts… (read_file) agent: Found typo at line 47. Fixing… (edit_file) agent: Running tests… (run_bash) agent: Opening PR via flukebase_pr_create…
Walks cwd, resolves the FlukeBase project, injects context. The agent knows your tech stack, memory, and deploys before the first prompt.
Reads your ~/.claude.json on startup. Get hosted git, CI/CD, payments, email, meeting AI — every tool the operator exposes.
Every conversation persists to disk. flukeclaw chat -c picks up where you left off — even after a reboot.
FlukeClaw is built in the open inside the FlukeBase monorepo. Each row is a merged PR.
Path-sandboxed. Soft-failure on non-zero exits so the agent reads stderr and recovers.
Speaks JSON-RPC 2.0 streamable HTTP. Reads bearer tokens from ~/.claude.json. Multi-server registry with collision-safe naming.
Append-only JSONL at ~/.flukeclaw/sessions/. --continue resumes the most recent. flukeclaw sessions lists them.
Resolves cwd → flukebase_get_project, injects project context into the system prompt, defaults project_id on every flukebase_* tool call.
brew install flukeclaw
Drop your email. We'll ping you the moment brew install flukeclaw works.
No spam. One email when it ships. Unsubscribe with one click.