lean-lsp-mcp 0.11.1__py3-none-any.whl → 0.11.3__py3-none-any.whl

lean_lsp_mcp/client_utils.py

@@ -64,59 +64,75 @@ def valid_lean_project_path(path: Path | str) -> bool:
     return (path_obj / "lean-toolchain").is_file()
 
 
-def setup_client_for_file(ctx: Context, file_path: str) -> str | None:
-    """Ensure the LSP client matches the file's Lean project and return its relative path."""
+def infer_project_path(ctx: Context, file_path: str) -> Path | None:
+    """Infer and cache the Lean project path for a file WITHOUT starting the client.
+
+    Walks up the directory tree to find a lean-toolchain file, caches the result.
+    Sets ctx.request_context.lifespan_context.lean_project_path if found.
+
+    Side effects when path changes:
+    - Next LSP tool will restart the client for the new project
+    - File content hashes will be cleared
 
+    Args:
+        ctx (Context): Context object
+        file_path (str): Absolute or relative path to a Lean file
+
+    Returns:
+        Path | None: The resolved project path if found, None otherwise
+    """
     lifespan = ctx.request_context.lifespan_context
-    project_cache = getattr(lifespan, "project_cache", {})
     if not hasattr(lifespan, "project_cache"):
-        lifespan.project_cache = project_cache
+        lifespan.project_cache = {}
 
     abs_file_path = os.path.abspath(file_path)
     file_dir = os.path.dirname(abs_file_path)
 
-    def activate_project(project_path: Path, cache_dirs: list[str]) -> str | None:
-        project_path_obj = project_path
-        rel = get_relative_file_path(project_path_obj, file_path)
-        if rel is None:
+    def set_project_path(project_path: Path, cache_dirs: list[str]) -> Path | None:
+        """Validate file is in project, set path, update cache."""
+        if get_relative_file_path(project_path, file_path) is None:
             return None
 
-        project_path_obj = project_path_obj.resolve()
-        lifespan.lean_project_path = project_path_obj
+        project_path = project_path.resolve()
+        lifespan.lean_project_path = project_path
 
-        cache_targets: list[str] = []
-        for directory in cache_dirs + [str(project_path_obj)]:
-            if directory and directory not in cache_targets:
-                cache_targets.append(directory)
+        # Update all relevant directories in cache
+        for directory in set(cache_dirs + [str(project_path)]):
+            if directory:
+                lifespan.project_cache[directory] = project_path
 
-        for directory in cache_targets:
-            project_cache[directory] = project_path_obj
+        return project_path
 
-        startup_client(ctx)
-        return rel
+    # Fast path: current project already valid for this file
+    if lifespan.lean_project_path and set_project_path(
+        lifespan.lean_project_path, [file_dir]
+    ):
+        return lifespan.lean_project_path
 
-    # Fast path: current Lean project already valid for this file
-    if lifespan.lean_project_path is not None:
-        rel_path = activate_project(lifespan.lean_project_path, [file_dir])
-        if rel_path is not None:
-            return rel_path
-
-    # Walk up from file directory to root, using cache hits or lean-toolchain
-    prev_dir = None
+    # Walk up directory tree using cache and lean-toolchain detection
     current_dir = file_dir
-    while current_dir and current_dir != prev_dir:
-        cached_root = project_cache.get(current_dir)
+    while current_dir and current_dir != os.path.dirname(current_dir):
+        cached_root = lifespan.project_cache.get(current_dir)
+
         if cached_root:
-            rel_path = activate_project(Path(cached_root), [current_dir])
-            if rel_path is not None:
-                return rel_path
+            if result := set_project_path(Path(cached_root), [current_dir]):
+                return result
         elif valid_lean_project_path(current_dir):
-            rel_path = activate_project(Path(current_dir), [current_dir])
-            if rel_path is not None:
-                return rel_path
+            if result := set_project_path(Path(current_dir), [current_dir]):
+                return result
         else:
-            project_cache[current_dir] = ""
-        prev_dir = current_dir
+            lifespan.project_cache[current_dir] = ""  # Mark as checked
+
         current_dir = os.path.dirname(current_dir)
 
     return None
+
+
+def setup_client_for_file(ctx: Context, file_path: str) -> str | None:
+    """Ensure the LSP client matches the file's Lean project and return its relative path."""
+    project_path = infer_project_path(ctx, file_path)
+    if project_path is None:
+        return None
+
+    startup_client(ctx)
+    return get_relative_file_path(project_path, file_path)

lean_lsp_mcp/search_utils.py

@@ -56,10 +56,10 @@ def lean_local_search(
 ) -> list[dict[str, str]]:
     """Search Lean declarations matching ``query`` using ripgrep; results include theorems, lemmas, defs, classes, instances, structures, inductives, abbrevs, and opaque decls."""
     root = (project_root or Path.cwd()).resolve()
-    escaped_query = re.escape(query)
-    ripgrep_pattern = (
+
+    pattern = (
         rf"^\s*(?:theorem|lemma|def|axiom|class|instance|structure|inductive|abbrev|opaque)\s+"
-        rf"{escaped_query}[A-Za-z0-9_'.]*(?:\s|:)"
+        rf"(?:[A-Za-z0-9_'.]+\.)*{re.escape(query)}[A-Za-z0-9_'.]*(?:\s|:)"
     )
 
     command = [
@@ -70,68 +70,55 @@ def lean_local_search(
         "--hidden",
         "--color",
         "never",
+        "--no-messages",
         "-g",
         "*.lean",
         "-g",
         "!.git/**",
         "-g",
         "!.lake/build/**",
-        ripgrep_pattern,
-        ".",
+        pattern,
+        str(root),
     ]
 
-    lean_src_path = _get_lean_src_search_path()
-    if lean_src_path is not None:
-        command.append(lean_src_path)
-
-    completed = subprocess.run(
-        command,
-        capture_output=True,
-        text=True,
-        cwd=str(root),
-    )
+    if lean_src := _get_lean_src_search_path():
+        command.append(lean_src)
 
-    results: list[dict[str, str]] = []
+    result = subprocess.run(command, capture_output=True, text=True, cwd=str(root))
 
-    for raw_line in completed.stdout.splitlines():
-        if not raw_line:
-            continue
-
-        event = _json_loads(raw_line)
-
-        if event.get("type") != "match":
+    matches = []
+    for line in result.stdout.splitlines():
+        if not line or (event := _json_loads(line)).get("type") != "match":
             continue
 
         data = event["data"]
-        line_text = data["lines"]["text"]
-        parts = line_text.lstrip().split(maxsplit=2)
+        parts = data["lines"]["text"].lstrip().split(maxsplit=2)
         if len(parts) < 2:
             continue
 
-        decl_kind, raw_name = parts[0], parts[1]
-        decl_name = raw_name.rstrip(":")
-
-        path_text = data["path"]["text"]
-        file_path = Path(path_text)
-        absolute_path = (
+        decl_kind, decl_name = parts[0], parts[1].rstrip(":")
+        file_path = Path(data["path"]["text"])
+        abs_path = (
             file_path if file_path.is_absolute() else (root / file_path).resolve()
         )
+
         try:
-            display_path = str(absolute_path.relative_to(root))
+            display_path = str(abs_path.relative_to(root))
         except ValueError:
             display_path = str(file_path)
 
-        results.append({"name": decl_name, "kind": decl_kind, "file": display_path})
+        matches.append({"name": decl_name, "kind": decl_kind, "file": display_path})
 
-        if len(results) >= limit:
+        if len(matches) >= limit:
             break
 
-    if completed.returncode not in (0, 1):
-        raise RuntimeError(
-            f"ripgrep exited with code {completed.returncode}\n{completed.stderr}"
-        )
+    if result.returncode not in (0, 1) and not matches:
+        error_msg = f"ripgrep exited with code {result.returncode}"
+        if result.stderr:
+            error_msg += f"\n{result.stderr}"
+        raise RuntimeError(error_msg)
 
-    return results
+    return matches
 
 
 @lru_cache(maxsize=1)

lean_lsp_mcp/server.py

@@ -18,7 +18,11 @@ from mcp.server.fastmcp.utilities.logging import get_logger, configure_logging
 from mcp.server.auth.settings import AuthSettings
 from leanclient import LeanLSPClient, DocumentContentChange
 
-from lean_lsp_mcp.client_utils import setup_client_for_file, startup_client
+from lean_lsp_mcp.client_utils import (
+    setup_client_for_file,
+    startup_client,
+    infer_project_path,
+)
 from lean_lsp_mcp.file_utils import get_file_contents, update_file
 from lean_lsp_mcp.instructions import INSTRUCTIONS
 from lean_lsp_mcp.search_utils import check_ripgrep_status, lean_local_search
@@ -154,10 +158,7 @@ async def lsp_build(
         ctx.request_context.lifespan_context.lean_project_path = lean_project_path_obj
 
     if lean_project_path_obj is None:
-        return (
-            "Lean project path not known yet. Provide `lean_project_path` explicitly or call a "
-            "tool that infers it (e.g. `lean_goal`) before running `lean_build`."
-        )
+        return "Lean project path not known yet. Provide `lean_project_path` explicitly or call a tool that infers it (e.g. `lean_file_contents`) before running `lean_build`."
 
     build_output = ""
     try:
@@ -247,6 +248,10 @@ def file_contents(ctx: Context, file_path: str, annotate_lines: bool = True) ->
     Returns:
         str: File content or error msg
     """
+    # Infer project path but do not start a client
+    if file_path.endswith(".lean"):
+        infer_project_path(ctx, file_path)  # Silently fails for non-project files
+
     try:
         data = get_file_contents(file_path)
     except FileNotFoundError:
@@ -572,9 +577,7 @@ def multi_attempt(
             # Apply the change to the file, capture diagnostics and goal state
             client.update_file(rel_path, [change])
             diag = client.get_diagnostics(rel_path)
-            formatted_diag = "\n".join(
-                format_diagnostics(diag, select_line=line - 1)
-            )
+            formatted_diag = "\n".join(format_diagnostics(diag, select_line=line - 1))
             # Use the snippet text length without any trailing newline for the column
             goal = client.get_goal(rel_path, line - 1, len(snippet_str))
             formatted_goal = format_goal(goal, "Missing goal")
@@ -585,7 +588,9 @@ def multi_attempt(
         try:
             client.close_files([rel_path])
         except Exception as exc:  # pragma: no cover - close failures only logged
-            logger.warning("Failed to close `%s` after multi_attempt: %s", rel_path, exc)
+            logger.warning(
+                "Failed to close `%s` after multi_attempt: %s", rel_path, exc
+            )
 
 
 @mcp.tool("lean_run_code")
@@ -604,7 +609,7 @@ def run_code(ctx: Context, code: str) -> List[str] | str:
     lifespan_context = ctx.request_context.lifespan_context
     lean_project_path = lifespan_context.lean_project_path
     if lean_project_path is None:
-        return "No valid Lean project path found. Run another tool (e.g. `lean_diagnostic_messages`) first to set it up or set the LEAN_PROJECT_PATH environment variable."
+        return "No valid Lean project path found. Run another tool (e.g. `lean_file_contents`) first to set it up."
 
     # Use a unique snippet filename to avoid collisions under concurrency
     rel_path = f"_mcp_snippet_{uuid.uuid4().hex}.lean"
@@ -683,11 +688,10 @@ def local_search(
         return _RG_MESSAGE
 
     stored_root = ctx.request_context.lifespan_context.lean_project_path
-    project_root = Path(stored_root) if stored_root else None
-    results = lean_local_search(
-        query=query.strip(), limit=limit, project_root=project_root
-    )
-    return results
+    if stored_root is None:
+        return "Lean project path not set. Call a file-based tool (like lean_file_contents) first to set the project path."
+
+    return lean_local_search(query=query.strip(), limit=limit, project_root=stored_root)
 
 
 @mcp.tool("lean_leansearch")
@@ -711,9 +715,7 @@ def leansearch(ctx: Context, query: str, num_results: int = 5) -> List[Dict] | s
     """
     try:
         headers = {"User-Agent": "lean-lsp-mcp/0.1", "Content-Type": "application/json"}
-        payload = orjson.dumps(
-            {"num_results": str(num_results), "query": [query]}
-        )
+        payload = orjson.dumps({"num_results": str(num_results), "query": [query]})
 
         req = urllib.request.Request(
             "https://leansearch.net/search",
@@ -784,73 +786,52 @@ def loogle(ctx: Context, query: str, num_results: int = 8) -> List[dict] | str:
 
 @mcp.tool("lean_leanfinder")
 @rate_limited("leanfinder", max_requests=10, per_seconds=30)
-def leanfinder(
-    ctx: Context, query: str, num_results: int = 5
-) -> List[tuple] | str:
-    """Search Mathlib theorems/definitions semantically by mathematical concept using Lean Finder.
+def leanfinder(ctx: Context, query: str, num_results: int = 5) -> List[Dict] | str:
+    """Search Mathlib theorems/definitions semantically by mathematical concept or proof state using Lean Finder.
 
     Effective query types:
-    - Math + API: "setAverage Icc interval", "integral_pow symmetric bounds"
-    - Conceptual: "algebraic elements same minimal polynomial", "quadrature nodes"
-    - Structure: "Finset expect sum commute", "polynomial degree bounded eval"
-    - Natural: "average equals point values", "root implies equal polynomials"
+    - Natural language mathematical statement: "For any natural numbers n and m, the sum n+m is equal to m+n."
+    - Natural language questions: "I'm working with algebraic elements over a field extension … Does this imply that the minimal polynomials of x and y are equal?"
+    - Proof state. For better results, enter a proof state followed by how you want to transform the proof state.
+    - Statement definition: Fragment or the whole statement definition.
 
-    Tips: Mix informal math terms with Lean identifiers. Multiple targeted queries beat one complex query.
+    Tips: Multiple targeted queries beat one complex query.
 
     Args:
-        query (str): Mathematical concepts combined with Lean terms
+        query (str): Mathematical concept or proof state
         num_results (int, optional): Max results. Defaults to 5.
 
     Returns:
-        List[tuple] | str: (lean_statement, english_description) pairs or error
+        List[Dict] | str: List of Lean statement objects (full name, formal statement, informal statement) or error msg
     """
     try:
         headers = {"User-Agent": "lean-lsp-mcp/0.1", "Content-Type": "application/json"}
-        payload = orjson.dumps({"data": [query, num_results, "Normal"]})
-
+        request_url = (
+            "https://bxrituxuhpc70w8w.us-east-1.aws.endpoints.huggingface.cloud"
+        )
+        payload = orjson.dumps({"inputs": query, "top_k": int(num_results)})
         req = urllib.request.Request(
-            "https://delta-lab-ai-lean-finder.hf.space/gradio_api/call/retrieve",
-            data=payload,
-            headers=headers,
-            method="POST",
+            request_url, data=payload, headers=headers, method="POST"
         )
 
-        with urllib.request.urlopen(req, timeout=10) as response:
-            event_data = orjson.loads(response.read())
-            event_id = event_data.get("event_id")
-
-        if not event_id:
-            return "Lean Finder has timed out or errored. It might be warming up, try a second time in 2 minutes."
-
-        result_url = f"https://delta-lab-ai-lean-finder.hf.space/gradio_api/call/retrieve/{event_id}"
-        req = urllib.request.Request(result_url, headers=headers, method="GET")
-
+        results = []
         with urllib.request.urlopen(req, timeout=30) as response:
-            for line in response:
-                line = line.decode("utf-8").strip()
-                if line.startswith("data: "):
-                    data = orjson.loads(line[6:])
-                    if isinstance(data, list) and len(data) > 0:
-                        html = data[0] if isinstance(data[0], str) else str(data)
-
-                        # Parse HTML table rows
-                        rows = re.findall(
-                            r"<tr><td>\d+</td><td>(.*?)</td><td>(.*?)</td></tr>",
-                            html, re.DOTALL
-                        )
-                        results = []
-                        for formal_cell, informal_cell in rows:
-                            formal = re.search(r"<code[^>]*>(.*?)</code>", formal_cell, re.DOTALL)
-                            informal = re.search(r"<span[^>]*>(.*?)</span>", informal_cell, re.DOTALL)
-                            if formal:
-                                results.append((
-                                    formal.group(1).strip(),
-                                    informal.group(1).strip() if informal else ""
-                                ))
-
-                        return results if results else "Lean Finder: No results parsed"
-
-        return "Lean Finder: No results received"
+            data = orjson.loads(response.read())
+            for result in data["results"]:
+                if (
+                    "https://leanprover-community.github.io/mathlib4_docs"
+                    not in result["url"]
+                ):  # Do not include results from other sources other than mathlib4, since users might not have imported them
+                    continue
+                full_name = re.search(r"pattern=(.*?)#doc", result["url"]).group(1)
+                obj = {
+                    "full_name": full_name,
+                    "formal_statement": result["formal_statement"],
+                    "informal_statement": result["informal_statement"],
+                }
+                results.append(obj)
+
+        return results if results else "Lean Finder: No results parsed"
     except Exception as e:
         return f"Lean Finder Error:\n{str(e)}"
 

{lean_lsp_mcp-0.11.1.dist-info → lean_lsp_mcp-0.11.3.dist-info}/METADATA

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: lean-lsp-mcp
-Version: 0.11.1
+Version: 0.11.3
 Summary: Lean Theorem Prover MCP
 Author-email: Oliver Dressler <hey@oli.show>
 License-Expression: MIT
@@ -543,6 +543,10 @@ uv sync --all-extras
 uv run pytest tests
 ```
 
+## Publications using lean-lsp-mcp
+
+- Ax-Prover: A Deep Reasoning Agentic Framework for Theorem Proving in Mathematics and Quantum Physics [arxiv](https://arxiv.org/abs/2510.12787)
+
 ## Related Projects
 
 - [LeanTool](https://github.com/GasStationManager/LeanTool)

lean_lsp_mcp-0.11.3.dist-info/RECORD

@@ -0,0 +1,14 @@
+lean_lsp_mcp/__init__.py,sha256=lxqDq0G_sI2iu2Nniy-pTW7BE9Ux7ZXeDoGf0OAWIDc,763
+lean_lsp_mcp/__main__.py,sha256=XnpTzfJc0T-j9tHtdkA8ovTr1c139ffTewcJGhxYDaM,49
+lean_lsp_mcp/client_utils.py,sha256=FNfiEWQagRj9tmo2xUDxbSZYVz5_RfaNH6OApgNQ9ZM,5065
+lean_lsp_mcp/file_utils.py,sha256=qddegF-T5-egZop8dPe_3Cma-3rRSKsAErVDQLecmbE,2916
+lean_lsp_mcp/instructions.py,sha256=y_gHlbeJoKnPohmcSVrQQds6mbBO1en-lxnXAfEypZE,892
+lean_lsp_mcp/search_utils.py,sha256=X2LPynDNLi767UDxbxHpMccOkbnfKJKv_HxvRNxIXM4,3984
+lean_lsp_mcp/server.py,sha256=zwiHeMlgRpERNIF-E_6OKzI8a-YmAx09EVX3QO2gJAk,34792
+lean_lsp_mcp/utils.py,sha256=zLu2VIhaX4yocY07F3Z94LB2jRGrkH1ID9SjR3poE9A,8255
+lean_lsp_mcp-0.11.3.dist-info/licenses/LICENSE,sha256=CQlxnf0tQyoVrBE93JYvAUYxv6Z5Yg6sX0pwogOkFvo,1071
+lean_lsp_mcp-0.11.3.dist-info/METADATA,sha256=1mf98hyaVVN1lZLWMZEdtSV6-nDZ7KAFyVLPRc7gbx8,19626
+lean_lsp_mcp-0.11.3.dist-info/WHEEL,sha256=_zCd3N1l69ArxyTb8rzEoP9TpbYXkqRFSNOD5OuxnTs,91
+lean_lsp_mcp-0.11.3.dist-info/entry_points.txt,sha256=nQbvwctWkWD7I-2f4VrdVQBZYGUw8CnUnFC6QjXxOSE,51
+lean_lsp_mcp-0.11.3.dist-info/top_level.txt,sha256=LGEK0lgMSNPIQ6mG8EO-adaZEGPi_0daDs004epOTF0,13
+lean_lsp_mcp-0.11.3.dist-info/RECORD,,

lean_lsp_mcp-0.11.1.dist-info/RECORD

@@ -1,14 +0,0 @@
-lean_lsp_mcp/__init__.py,sha256=lxqDq0G_sI2iu2Nniy-pTW7BE9Ux7ZXeDoGf0OAWIDc,763
-lean_lsp_mcp/__main__.py,sha256=XnpTzfJc0T-j9tHtdkA8ovTr1c139ffTewcJGhxYDaM,49
-lean_lsp_mcp/client_utils.py,sha256=Z-2AoFBA36UO8oeauTNUtquE6Yz_ZRTKz_9n1iapyJ4,4519
-lean_lsp_mcp/file_utils.py,sha256=qddegF-T5-egZop8dPe_3Cma-3rRSKsAErVDQLecmbE,2916
-lean_lsp_mcp/instructions.py,sha256=y_gHlbeJoKnPohmcSVrQQds6mbBO1en-lxnXAfEypZE,892
-lean_lsp_mcp/search_utils.py,sha256=cNpWzR1TuNyvRxCeCIb0HoEHGsCIW1ypqN1R1ZpbkdI,4170
-lean_lsp_mcp/server.py,sha256=jK2bwi7iiM_agXtwRNFAMIpAo5FLDJ2FKfdDLvbaWfY,35433
-lean_lsp_mcp/utils.py,sha256=zLu2VIhaX4yocY07F3Z94LB2jRGrkH1ID9SjR3poE9A,8255
-lean_lsp_mcp-0.11.1.dist-info/licenses/LICENSE,sha256=CQlxnf0tQyoVrBE93JYvAUYxv6Z5Yg6sX0pwogOkFvo,1071
-lean_lsp_mcp-0.11.1.dist-info/METADATA,sha256=IWxzDA2qqh5yK7jxCr36R6z7HuwwLV2bzoFlGqUWEoQ,19444
-lean_lsp_mcp-0.11.1.dist-info/WHEEL,sha256=_zCd3N1l69ArxyTb8rzEoP9TpbYXkqRFSNOD5OuxnTs,91
-lean_lsp_mcp-0.11.1.dist-info/entry_points.txt,sha256=nQbvwctWkWD7I-2f4VrdVQBZYGUw8CnUnFC6QjXxOSE,51
-lean_lsp_mcp-0.11.1.dist-info/top_level.txt,sha256=LGEK0lgMSNPIQ6mG8EO-adaZEGPi_0daDs004epOTF0,13
-lean_lsp_mcp-0.11.1.dist-info/RECORD,,