v2.0.0

Breaking

• Renamed the installed package from src to agent_spawn_mcp. The agent-spawn-mcp CLI script and uvx agent-spawn-mcp invocations are unchanged. Code that imported from src.* against this package must update to from agent_spawn_mcp.*.

Fixed

• load_dotenv("example.env") no longer overrides real .env files. Both the CLI entry point and the config loader now call load_dotenv() so the user's .env (created via cp example.env .env) is actually loaded.
• list_chat_sessions no longer raises KeyError on histories that were written before timestamps were recorded; missing time falls back to "?". New entries written by save_history always include a time field.
• Removed the __import__("base64") hack in agent_spawn_mcp.utils; the module imports base64 once at the top level.

Changed

• OpenAICompatProvider now keeps one httpx.Client per instance and the spawn-agent factory reuses a single provider across calls, so HTTP connections are pooled instead of opened and closed per request. Per-call timeout is applied at request level.
• web_search (full server) and the agent tool now forward allowed_domains / excluded_domains into the underlying web_search tool payload. Both raise on mutually exclusive arguments.
• BaseProvider abstract return types now reflect the actual implementation (dict / bytes) instead of misleading httpx.Response. The unused default _request helper has been removed.
• Added the Programming Language :: Python :: 3.13 trove classifier and added Python 3.13 to the CI test matrix.

v1.1.3

Changed

• Refreshed the lockfile to the latest compatible dependency versions.
• Updated GitHub Actions to use locked dependency installs for tests and to
  check the lockfile before publishing.
• Added --token-env for spawn-agent MCP servers so persistent client configs
  can reference token environment variables.
• Added README instructions for registering AgentSpawnMCP in Codex CLI.
• Added a README hero image.

v1.1.2

Changelog

1.1.2 - 2026-05-14

Fixed

• Fixed spawn-agent response extraction for providers that return text content
  blocks instead of a plain string.
• Fixed full-server --url and --local startup so inline providers are not
  overwritten by a second YAML config load.
• Fixed full-server --model override so it is applied after the active
  provider is loaded.
• Fixed the list_providers MCP tool by avoiding a helper/function name
  collision.

Changed

• Added early validation for empty spawn-agent tasks, missing models, invalid
  timeouts, invalid max_tokens, and unsupported API types.
• Restricted the CLI --api-type option to openai or anthropic.
• Expanded AGENTS.md into a more complete developer guide with provider
  loading rules, tool patterns, validation expectations, and release-relevant
  gotchas.

Tests

• Added coverage for spawn-agent metadata, text-block normalization, input
  validation, inline providers, model overrides, list_providers, and API type
  validation.

Release 1.1.1

What's Changed

• Release 1.1.0: fix URL versioning; add z.ai provider by @sandsaber in #1
• Remove client-side max_tokens cap (1.1.1) by @sandsaber in #2

New Contributors

• @sandsaber made their first contribution in #1

Full Changelog: v1.1.0...v1.1.1

Release 1.1.0

Full Changelog: v1.0.5...v1.1.0

Release 1.0.5

Full Changelog: v1.0.4...v1.0.5

Release 1.0.4

Full Changelog: v1.0.3...v1.0.4