pi-claude-agent-sdk

A provider-only fork of pi-claude-bridge, synchronized through upstream v0.12.0. This fork uses Pi's Anthropic authentication, defaults to the Max plan, and does not register AskClaude.

Use Claude models in pi with your Claude subscription.

Is this allowed? Yes, as far as I can tell. Anthropic at one point announced and then unannounced a billing change for Agent SDK tools like this one. Things could change again, but right now the bridge uses your regular subscription quota just like Claude Code.

[!NOTE] There's another good option: the pi maintainers now have their own Claude Code subscription provider, pi-claude-code. The two work differently under the hood:

• pi-claude-agent-sdk (this fork) runs Claude Code through the Agent SDK and keeps pi's message history in sync with a Claude Code session file on disk. Claude Code builds and sends every request itself, unmodified, with its own system prompt and pi's prompt sections added to it. It tries to stay true to how Claude Code normally runs.
  • Tradeoff: complexity. Other extensions that change the system prompt mid-session, add messages to requests, or run subagents with their own prompts can fail.
• pi-claude-code runs claude -p for each turn and points it at a local proxy that rewrites each request on its way to Anthropic: it swaps in pi's history and system prompt as-is and adjusts cache markers. It keeps no state of its own, so it's simpler and gets along better with other extensions.
  • Tradeoff: editing inference requests on the wire is less clearly within the TOS, and Claude Code updates may introduce breaking changes without warning.

Install

The bridge needs pi 1.5.1 or newer.

1. Install the extension:

   pi install npm:pi-claude-agent-sdk

2. Configure Anthropic authentication in Pi with /login (Claude subscription OAuth) or an Anthropic API key. Claude Code login is not used as a fallback. Before each new subprocess, Pi resolves the credential and refreshes OAuth early when fewer than two hours remain.
3. In pi, pick a claude-bridge/... model with /model, e.g. claude-bridge/claude-fable-5-2. The provider ID stays claude-bridge for compatibility; anthropic/... models use Pi's built-in provider instead.

Usage

The model list comes from pi-ai's Anthropic catalog automatically. When pi-ai adds a new Claude model, it appears in /model after you update the package; no bridge update is needed. Dated snapshot ids (e.g. claude-opus-4-5-20251101) are not shown.

All current models run with 1M context. Opus 4.6 and Sonnet 4.6 need extra settings for 1M (see Configuration).

pi's tools are bridged to Claude Code, but you continue to use pi as usual:

• Bash commands get Claude Code's 120-second default timeout since Claude expects it but pi's bash has no default limit.
• Skills are forwarded to Claude Code's system prompt.
• Steering mid-turn works and reaches Claude at the next tool boundary.

Configuration

Create ~/.pi/agent/claude-bridge.json (or under PI_CODING_AGENT_DIR). Project-level provider configuration is ignored for security.

{
  "provider": {
    "strictMcpConfig": true,
    "pathToClaudeCodeExecutable": "/home/you/.nix-profile/bin/claude"
  }
}

Provider options

• strictMcpConfig - block MCP servers from ~/.claude.json / .mcp.json (default true). Cloud MCP (Gmail/Drive via claude.ai OAuth) is always blocked.
• autoMemoryEnabled - enable Claude Code's auto-memory system (default false).
• pathToClaudeCodeExecutable - path to the claude binary. Useful if your OS/filesystem has the SDK's bundled musl/glibc binaries in a place where they can't run. For example, with Nix you can set the binary to "/home/you/.nix-profile/bin/claude".
• forceTwoHundredK - array of model ids to pin to 200K context (bare id, no [1m] suffix). Use if pi-ai declares a model at 1M but Claude Code won't serve it on your plan.
• plan - defaults to "max", enabling Opus 4.6 with 1M context. Set "pro" on a Pro plan to keep that model at 200K.
• longContextExtraUsage - set to true to run Opus 4.6 (on Pro) and Sonnet 4.6 with 1M context, billed through Extra Usage.

Model overrides

pi's modelOverrides in ~/.pi/agent/models.json do not work correctly with pi-claude-bridge; they only apply to pi-side settings like compaction threshold. Overriding contextWindow or other fields requires editing src/models.ts directly. To pin a model to 200K, use provider.forceTwoHundredK instead.

Environment variables

The child inherits non-authentication environment variables. Pi's resolved Anthropic credential, base URL and headers replace inherited Claude/Anthropic authentication and alternate backend settings; raw auth variables cannot override the credential Pi selected. Bedrock, Vertex and Foundry are not supported by this fork.

To set non-authentication variables only for bridge requests (for example a proxy), add an env object to an anthropic credential or a claude-bridge placeholder credential in ~/.pi/agent/auth.json:

• Values override pi's environment.
• An empty value removes the variable.
• Extensions calling a bridge model can pass env per request the same way.

CLAUDE_CONFIG_DIR, HOME and USERPROFILE can't be set there (the turn fails), since the bridge reads Claude Code's session files through pi's own values.

This example sends bridge requests through a proxy. The placeholder does not authenticate the child; the anthropic credential configured in Pi is still required:

{
  "claude-bridge": {
    "type": "api_key",
    "key": "not-used",
    "env": {
      "HTTPS_PROXY": "http://127.0.0.1:8080"
    }
  }
}

Session history

The bridge compares conversation content before each fresh query:

• Unchanged history resumes the Claude Code session.
• Tree navigation and history edits retain Claude Code's records up to the last safe matching point, then continue in a new session. /fork and clone start from the parent session's Claude Code records the same way.
• Model changes and changed instructions import pi's history into a fresh Claude Code session.

Session pointers, content digests and record IDs are saved under claude-bridge/sessions/ in pi's agent directory (PI_CODING_AGENT_DIR, default ~/.pi/agent). They let the bridge recover its matching prefix after a restart. Running the same pi session simultaneously in multiple bridge processes is unsupported.

Compatibility with other extensions

Which injection routes reach Claude Code?

The bridge forwards pi's structured parts: project context files, skills, custom prompt, appended instructions, and custom prompt sections (systemPromptOptions.sections). It drops the rest. Measured against the request body (diag/capture-proxy.mjs):

Two things to watch for:

• Returning systemPrompt from before_agent_start makes pi replace the whole system prompt, discarding any context_with_system edit in the same run. Only one reaches the request.
• System-prompt edits work by wrapping. Replacing pi's prompt leaves the bridge with nothing to match, so it refuses the turn rather than send Claude Code a request missing your context files, skills and custom instructions. The error names the closest known prompt and where it diverged.

To add instructions, use message, context, or a system-message edit that keeps pi's prompt intact.

Extensions that add a message to every request

Some extensions add a message from their context hook on every request without saving it to pi's history (pi-todo-rail does this with its todo list). Claude Code keeps each copy and puts its cache marker on it, but the next request no longer has that copy. Nothing cached after the system prompt can be reused: every request rewrites the whole prompt cache, and Claude Code restarts after every tool call. The bridge warns after three such requests in a row and quotes the message.

To fix it, disable the extension, or have it save the message with pi.sendMessage when its content changes so it stays in pi's history.

Hooks written for Claude Code

~/.claude/settings.json hooks fire inside bridge turns, so a hook injecting Claude-specific guidance duplicates what pi's extensions already provide. A project's .claude/settings.json and .claude/settings.local.json are not loaded, so a repo's hooks, apiKeyHelper and env never run.

pi sets PI_CODING_AGENT=true for child processes, including the Claude Code child. A hook can skip itself on that:

[ -n "$PI_CODING_AGENT" ] && exit 0

Hooks do not fire on isolated summary queries.

Compaction and branch summaries

pi runs compaction and branch summaries natively. The bridge routes their cacheRetention: "none" calls to isolated Claude Code queries, leaving the live session untouched until pi completes the history rewrite. Other extensions can provide summaries through pi's hooks. Tool-free extension-owned modelRegistry.complete() calls marked cacheRetention: "none" use the same isolated path, honoring their private system prompt, reasoning level and usage without touching the chat session.

Extra Usage with plan limits left

Anthropic reads some system prompts as a third-party app's and bills those requests to Extra Usage even when your plan has room, or refuses them with "You're out of extra usage" when Extra Usage is off. pi's own prompt before pi 1.5 was one. The bridge warns when Claude Code reports a request billed to Extra Usage with no plan limit reached, and adds the same note to an "out of extra usage" error. Models you run at 1M through longContextExtraUsage are left out, since they bill Extra Usage by design. If you see it, look for an extension that adds text to the system prompt.

Plan usage for footers and quota extensions

The bridge does not show plan usage itself. Claude Code sends a rate_limit_event when a usage window changes, and the bridge passes it to pi's after_provider_response event as anthropic-ratelimit-unified-* headers, using the Anthropic API's names (for example anthropic-ratelimit-unified-5h-utilization).

Events are not tied to one response. One that arrives after a turn has ended on a tool call is reported on the next provider call, and only the latest is kept. The bridge includes the headers it can make from the event: status, reset time, representative claim, overage status and reset, and utilization and reset for each window. The status code is always 200.

An extension that reads these headers for the built-in anthropic provider can also accept claude-bridge without other changes.

Using with @gotgenes/pi-subagents

Requires the following in ~/.pi/agent/subagents.json:

{"promptInheritance": {"claude-bridge": "portable"}}

Known issues

Edits far back in history can still cost cache

The unchanged prefix is retained, but Anthropic's cache lookup has a limited lookback. A long rewritten suffix can prevent reuse of an earlier cache entry.

File snapshots in rewritten history can be lost

Native records before the matching point keep their attachments. A rewritten suffix carries @file expansions, but not snapshots Claude Code recorded after editing files; the tool calls and results remain in pi's history.

Changed instructions cost a cache miss

When project context (AGENTS.md/CLAUDE.md), skills, or extension-written instructions change (for example, after an edit and /reload), the next turn imports the conversation into a new Claude Code session so the model sees them. That turn re-sends the conversation uncached and drops earlier thinking. An extension that rewrites its system prompt section every turn pays this on every turn.

Debugging

Set CLAUDE_BRIDGE_DEBUG=1 to enable debug output:

• Bridge log at claude-bridge.log in pi's agent dir (PI_CODING_AGENT_DIR, default ~/.pi/agent): provider calls, session sync decisions, tool results, CC stderr. Override location with CLAUDE_BRIDGE_DEBUG_PATH.
• Per-query CC CLI logs at cc-cli-logs/<timestamp>-<tag>-<seq>.log in the same directory: the subprocess's own debug stream; tag is provider or compact-summary. Shows CC's view of session loading, API requests, and tool calls.

When filing a bug about a session-resume failure (e.g. "No conversation found"), the most useful attachments are the syncResult: lines from the bridge log plus the matching cc-cli-logs/ file for the failing query.

Development

Tests

npm run test:unit for offline tests and npm run typecheck for TypeScript checks. npm test adds integration tests that hit APIs using Pi's Anthropic credential; set CLAUDE_BRIDGE_TESTING_ALT_PROVIDER and CLAUDE_BRIDGE_TESTING_ALT_MODEL in .env.test for cross-provider tests.

Integration tests spawn real pi and Claude Code subprocesses and need write access to ~/.claude -- a sandbox that blocks it makes --resume fail with No conversation found with session ID.

Acknowledgments

Originally based on claude-agent-sdk-pi by Prateek Sunal.