PerSQL — editor & agent plugins

Thin pointers that wire a coding agent into PerSQL. Two products, two folders:

• database/ — give the agent its own SQLite database

(create tables, run SQL, branch a sandbox, publish endpoints). Points at the remote mcp.persql.com server; the agent signs in (or signs up) via OAuth in the browser on first use — no API key to copy.

• context/ — give the agent a **shared, structured context

store** so memory follows you across Claude Code, OpenCode, Codex, and cloud. Runs through the local persql mcp stdio shim after persql login.

| Product | Claude Code | Codex | OpenCode | |---|---|---|---| | Database | database/claude | database/codex | database/opencode | | Context | context/claude | context/codex | context/opencode |

Database (database/)

# Claude Code
claude mcp add --transport http persql https://mcp.persql.com/mcp
# Codex
codex mcp add persql --url https://mcp.persql.com/mcp
# OpenCode — add the block from database/opencode/opencode.json

In Claude Code you can also install from the marketplace this repo defines:

/plugin marketplace add persql/plugins
/plugin install persql@persql            # database
/plugin install persql-context@persql    # shared context (needs the persql CLI)

The first tool call opens the browser to authorize. A brand-new user signs up right there and gets a primary workspace with a welcome credit; the client stores the credential, so it's a one-time step.

Context (context/)

npm install -g @persql/cli
persql login          # once
persql mcp            # stdio MCP server, proxies to context.persql.com

persql mcp resolves the context store automatically — per-repo (default, keyed on the git origin remote) or user-global (persql mcp --scope user). It injects the resolved store into every call, so the agent can't address the wrong one. Set PERSQL_TOKEN to a namespace token (psql_live_…).

Each context plugin also ships a short rule telling the agent to recall before starting and remember durable facts as it learns them.