-
[OPIK-4094] [SDK] Optimizer SDK reintroduce MCP tool use and optimization (#5056)
frostbyte_neo 发布于
2026-02-24 13:41:49 +00:00 -
refactor: mcp tool calling utils
-
refactor: restore prompt segment util
-
Update __init__.py
-
test: add mcp tests
-
Create test_mcp_remote_live.py
-
feat(mcp): update functions and sepc
-
docs: readme
-
feat: mcp tool calling libraries
-
chore: wiring
-
chore: update exampels
-
chore: update tests
-
chore: missing stubs
-
chore: mypy and state
-
chore: move prompts to util
-
chore: fixed schema validation for tool optimization
-
chore: mcp reporting display
-
fix: optimizer limits
-
Update toolcalling.py
-
chore: tests and formatters
-
chore: rename API and fix tests for mcp optimization
-
Update test_tool_factory.py
-
Update tool_factory.py
-
docs: tool optimization
-
Update test_prompt_segments.py
-
test: tool optimization
-
chore: tool optimization utils
-
Update tool_factory.py
-
chore: move tests
-
Update test_mcp_remote_live.py
-
feat: support tool optimization in evolutionary optimizer
-
chore: improve api for evo on tools
-
chore: evo toolops
-
feat: tool optimization few-shot
-
chore: api routing for tool optimization for evo
-
Update meta_prompt_optimizer.py
-
Update parameter_optimizer.py
-
Update base_optimizer.py
-
feat: tool optimization gepa + hrpo
-
chore: toolcalling
-
chore: lint
-
refactor: tool operations
-
chore: docstrings for tool use
-
Update mutation_ops.py
-
chore: docs updates
-
fix(optimizer): wire GEPA tool-use and role flag validation
-
fix(evolutionary): isolate child tool metadata and degrade on tool update errors
-
fix(agent): handle malformed tool-call arguments gracefully
-
fix(llm): enforce OpenAI strict structured-output schema rules
-
test(evolutionary): align semantic mutation fallback assertions
-
refactor(llm): centralize OpenAI response_format injection
-
fix(toolcalling): harden logging, selector mapping, and approval gates
-
refactor(display): unify tool summary rendering
-
chore(tests): satisfy pre-commit typing and formatting
-
fix(optimizer): degrade GEPA tool optimization and tighten flags
-
refactor(tests): mirror toolcalling test paths and remove chat_prompt duplication
-
docs(optimization): clarify MCP format mapping and GEPA tool limits
-
fix(ci): restore openai strict schema requirements
Move optional mcp SDK load test from unit to library_integration/mcp and remove broad exception wrapping in tool description candidate generation.
-
test(llm-calls): assert parsed structured output fields
-
feat(scripts): add remote MCP optimizer smoke scripts
-
chore(scripts): throttle remote MCP smoke runs
-
fix(security): remove sensitive identifiers from tool logs
-
chore(scripts): fail fast on optimizer smoke errors
-
refactor(scripts): centralize mcp tool description extraction
-
docs(scripts): clarify context7 metric behavior
-
fix(toolcalling): harden selector resolution and approval parsing
-
fix(api): preserve caller response_format and validate mcp mappings
-
refactor(toolcalling): split selector and rendering ops
-
feat(smoke): assert tool-signature optimization attempts
-
fix(toolcalling): align schema nullability and forward remote auth
-
feat(smoke): enforce single-item runs and prompt-change checks
-
fix(toolcalling): register mcp alias and close failed remote sessions
-
fix(smoke): keep context7 smoke dataset as Dataset instance
-
fix(reporting): add tqdm postfix compatibility for evaluator
-
feat(toolcalling): validate mcp output schemas and wire output overrides
-
chore(scripts): remove all-optimizer remote mcp smoke script
-
fix(reporting): drop duplicate tqdm set_postfix override
-
chore(opik-optimizer): remove unused npm manifest files
-
chore: various examples
-
python-sdk: source default evaluation model from OpikConfig
-
Revert "python-sdk: source default evaluation model from OpikConfig"
This reverts commit
b641b51df9.- Update sdks/opik_optimizer/README.md
Co-authored-by: baz-reviewer[bot] <174234987+baz-reviewer[bot]@users.noreply.github.com>
- Update sdks/opik_optimizer/scripts/optimize_user_only_example.py
Co-authored-by: baz-reviewer[bot] <174234987+baz-reviewer[bot]@users.noreply.github.com>
-
chore(scripts): apply pre-commit lint and formatting fixes
-
fix(few-shot): preserve multimodal content-part structure
-
test(few-shot): expand multimodal regression coverage
-
docs(optimization): align role-only examples with docs policy
-
fix(scripts): guard context7 metric against null model output
-
refactor(few-shot): extract multimodal-preservation helper and align docs/script logging
-
fix(few-shot): preserve non-empty multimodal text fallback
-
refactor(multimodal): centralize message-structure preservation helper
-
refactor(toolcalling): collapse component helpers and relocate tool description extraction
-
feat(api): accept cursor MCP config directly in ChatPrompt tools
-
chore(types): fix mypy annotations in scripts and tests
-
chore(examples): remove manual Cursor MCP conversion in user-facing usage
-
docs(readme): align cursor MCP tools usage with automatic ChatPrompt normalization
-
fix(multimodal): harden FSBO content fallback and strengthen tool tests
-
Update tool_optimization.mdx
-
Update generate_fern_docs.py
-
Update tool_optimization.mdx
-
Update prompt_segments_jinja_example.py
-
fix(python-sdk): keep v2 json encoders without pydantic.v1 imports
-
Revert "fix(python-sdk): keep v2 json encoders without pydantic.v1 imports"
This reverts commit
0796ae201c.-
docs(optimization): remove jinja role-enforcement examples
-
docs(readme): remove beta wording for MCP optimizer sections
Co-authored-by: baz-reviewer[bot] <174234987+baz-reviewer[bot]@users.noreply.github.com>
下载附件
-