langchain-mcp-tools 0.0.2__tar.gz → 0.0.4__tar.gz

This diff represents the content of publicly available package versions that have been released to one of the supported registries. The information contained in this diff is provided for informational purposes only and reflects changes between package versions as they appear in their respective public registries.
Files changed (15) hide show
  1. langchain_mcp_tools-0.0.4/PKG-INFO +172 -0
  2. langchain_mcp_tools-0.0.4/README.md +151 -0
  3. langchain_mcp_tools-0.0.4/langchain_mcp_tools.egg-info/PKG-INFO +172 -0
  4. {langchain_mcp_tools-0.0.2 → langchain_mcp_tools-0.0.4}/langchain_mcp_tools.egg-info/top_level.txt +0 -1
  5. {langchain_mcp_tools-0.0.2 → langchain_mcp_tools-0.0.4}/pyproject.toml +4 -6
  6. {langchain_mcp_tools-0.0.2 → langchain_mcp_tools-0.0.4}/src/langchain_mcp_tools.py +22 -12
  7. langchain_mcp_tools-0.0.2/PKG-INFO +0 -106
  8. langchain_mcp_tools-0.0.2/README.md +0 -83
  9. langchain_mcp_tools-0.0.2/langchain_mcp_tools.egg-info/PKG-INFO +0 -106
  10. {langchain_mcp_tools-0.0.2 → langchain_mcp_tools-0.0.4}/LICENSE +0 -0
  11. {langchain_mcp_tools-0.0.2 → langchain_mcp_tools-0.0.4}/langchain_mcp_tools.egg-info/SOURCES.txt +0 -0
  12. {langchain_mcp_tools-0.0.2 → langchain_mcp_tools-0.0.4}/langchain_mcp_tools.egg-info/dependency_links.txt +0 -0
  13. {langchain_mcp_tools-0.0.2 → langchain_mcp_tools-0.0.4}/langchain_mcp_tools.egg-info/requires.txt +0 -0
  14. {langchain_mcp_tools-0.0.2 → langchain_mcp_tools-0.0.4}/setup.cfg +0 -0
  15. {langchain_mcp_tools-0.0.2 → langchain_mcp_tools-0.0.4}/tests/test_langchain_mcp_tools.py +0 -0
langchain_mcp_tools-0.0.4/PKG-INFO ADDED
@@ -0,0 +1,172 @@
1
+ Metadata-Version: 2.2
2
+ Name: langchain-mcp-tools
3
+ Version: 0.0.4
4
+ Summary: Model Context Protocol (MCP) To LangChain Tools Conversion Utility
5
+ Project-URL: Bug Tracker, https://github.com/hideya/langchain-mcp-tools-py/issues
6
+ Project-URL: Source Code, https://github.com/hideya/langchain-mcp-tools-py
7
+ Requires-Python: >=3.11
8
+ Description-Content-Type: text/markdown
9
+ License-File: LICENSE
10
+ Requires-Dist: jsonschema-pydantic>=0.6
11
+ Requires-Dist: langchain-core>=0.3.29
12
+ Requires-Dist: langchain>=0.3.14
13
+ Requires-Dist: langchain-anthropic>=0.3.1
14
+ Requires-Dist: langchain-groq>=0.2.3
15
+ Requires-Dist: langchain-openai>=0.3.0
16
+ Requires-Dist: langgraph>=0.2.62
17
+ Requires-Dist: mcp>=1.2.0
18
+ Requires-Dist: pyjson5>=1.6.8
19
+ Requires-Dist: pympler>=1.1
20
+ Requires-Dist: python-dotenv>=1.0.1
21
+
22
+ # MCP To LangChain Tools Conversion Utility / Python [![License: MIT](https://img.shields.io/badge/License-MIT-blue.svg)](https://github.com/hideya/mcp-langchain-client-ts/blob/main/LICENSE) [![pypi version](https://img.shields.io/pypi/v/langchain-mcp-tools.svg)](https://pypi.org/project/langchain-mcp-tools/)
23
+
24
+ This package is intended to simplify the use of
25
+ [Model Context Protocol (MCP)](https://modelcontextprotocol.io/)
26
+ server tools with LangChain / Python.
27
+
28
+ It contains a utility function `convertMcpToLangchainTools()`.
29
+ This function handles parallel initialization of specified multiple MCP servers
30
+ and converts their available tools into an array of
31
+ [LangChain-compatible tools](https://js.langchain.com/docs/how_to/tool_calling/).
32
+
33
+ A typescript version of this utility library is available
34
+ [here](https://www.npmjs.com/package/@h1deya/langchain-mcp-tools)
35
+
36
+ ## Requirements
37
+
38
+ - Python 3.11+
39
+
40
+ ## Installation
41
+
42
+ ```bash
43
+ pip install langchain-mcp-tools
44
+ ```
45
+
46
+ ## Quick Start
47
+
48
+ `convert_mcp_to_langchain_tools()` utility function accepts MCP server configurations
49
+ that follow the same structure as
50
+ [Claude for Desktop](https://modelcontextprotocol.io/quickstart/user),
51
+ but only the contents of the `mcpServers` property,
52
+ and is expressed as a `dict`, e.g.:
53
+
54
+ ```python
55
+ mcp_configs = {
56
+ 'filesystem': {
57
+ 'command': 'npx',
58
+ 'args': ['-y', '@modelcontextprotocol/server-filesystem', '.']
59
+ },
60
+ 'fetch': {
61
+ 'command': 'uvx',
62
+ 'args': ['mcp-server-fetch']
63
+ }
64
+ }
65
+
66
+ tools, cleanup = await convert_mcp_to_langchain_tools(
67
+ mcp_configs
68
+ )
69
+ ```
70
+
71
+ The utility function initializes all specified MCP servers in parallel,
72
+ and returns LangChain Tools (`List[BaseTool]`)
73
+ by gathering all available MCP server tools,
74
+ and by wrapping them into [LangChain Tools](https://js.langchain.com/docs/how_to/tool_calling/).
75
+ It also returns a cleanup callback function (`McpServerCleanupFn`)
76
+ to be invoked to close all MCP server sessions when finished.
77
+
78
+ The returned tools can be used with LangChain, e.g.:
79
+
80
+ ```python
81
+ # from langchain.chat_models import init_chat_model
82
+ llm = init_chat_model(
83
+ model='claude-3-5-haiku-latest',
84
+ model_provider='anthropic'
85
+ )
86
+
87
+ # from langgraph.prebuilt import create_react_agent
88
+ agent = create_react_agent(
89
+ llm,
90
+ tools
91
+ )
92
+ ```
93
+ <!-- A simple and experimentable usage example can be found
94
+ [here](https://github.com/hideya/langchain-mcp-tools-ts-usage/blob/main/src/index.ts) -->
95
+
96
+ <!-- A more realistic usage example can be found
97
+ [here](https://github.com/hideya/langchain-mcp-client-ts) -->
98
+
99
+ An usage example can be found
100
+ [here](https://github.com/hideya/mcp-client-langchain-py)
101
+
102
+ ## Limitations
103
+
104
+ Currently, only text results of tool calls are supported.
105
+
106
+ ## Technical Details
107
+
108
+ It was very tricky (for me) to get the parallel MCP server initialization
109
+ to work successfully...
110
+
111
+ I'm new to Python, so it is very possible that my ignorance is playing
112
+ a big role here...
113
+ I'm summarizing why it was difficult for me below.
114
+ Any comments pointing out something I am missing would be greatly appreciated!
115
+ [(comment here)](https://github.com/hideya/langchain-mcp-tools-ts/issues)
116
+
117
+ 1. Core Challenge:
118
+ - Async resources management for `stdio_client` and `ClientSession` seems
119
+ to rely exclusively on `asynccontextmanager` for cleanup with no manual
120
+ cleanup options (based on the mcp python-sdk impl as of Jan 14, 2025)
121
+ - Initializing multiple MCP servers in parallel requires a dedicated
122
+ `asyncio.Task` per server
123
+ - Necessity of keeping sessions alive for later use after initialization
124
+ - Ensuring proper cleanup later in the same task that created them
125
+
126
+ 2. Solution Strategy:
127
+ A key requirement for parallel initialization is that each server must be
128
+ initialized in its own dedicated task - there's no way around this if we
129
+ want true parallel initialization. However, this creates a challenge since
130
+ we also need to maintain long-lived sessions and handle cleanup properly.
131
+
132
+ The key insight is to keep the initialization tasks alive throughout the
133
+ session lifetime, rather than letting them complete after initialization.
134
+
135
+ By using `asyncio.Event`s for coordination, we can:
136
+ - Allow parallel initialization while maintaining proper context management
137
+ - Keep each initialization task running until explicit cleanup is requested
138
+ - Ensure cleanup occurs in the same task that created the resources
139
+ - Provide a clean interface for the caller to manage the lifecycle
140
+
141
+ Alternative Considered:
142
+ A generator/coroutine approach using `finally` block for cleanup was
143
+ considered but rejected because:
144
+ - It turned out that the `finally` block in a generator/coroutine can be
145
+ executed by a different task than the one that ran the main body of
146
+ the code
147
+ - This breaks the requirement that `AsyncExitStack.aclose()` must be
148
+ called from the same task that created the context
149
+
150
+ 3. Task Lifecycle:
151
+ To allow the initialization task to stay alive waiting for cleanup:
152
+ ```
153
+ [Task starts]
154
+ ↓
155
+ Initialize server & convert tools
156
+ ↓
157
+ Set ready_event (signals tools are ready)
158
+ ↓
159
+ await cleanup_event.wait() (keeps task alive)
160
+ ↓
161
+ When cleanup_event is set:
162
+ exit_stack.aclose() (cleanup in original task)
163
+ ```
164
+ This approach indeed enables parallel initialization while maintaining proper
165
+ async resource lifecycle management through context managers.
166
+ However, I'm afraid I'm twisting things around too much.
167
+ It usually means I'm doing something very worng...
168
+
169
+ I think it is a natural assumption that MCP SDK is designed with consideration
170
+ for parallel server initialization.
171
+ I'm not sure what I'm missing...
172
+ (FYI, with the TypeScript SDK, parallel server initialization was pretty straightforward)
langchain_mcp_tools-0.0.4/README.md ADDED
@@ -0,0 +1,151 @@
1
+ # MCP To LangChain Tools Conversion Utility / Python [![License: MIT](https://img.shields.io/badge/License-MIT-blue.svg)](https://github.com/hideya/mcp-langchain-client-ts/blob/main/LICENSE) [![pypi version](https://img.shields.io/pypi/v/langchain-mcp-tools.svg)](https://pypi.org/project/langchain-mcp-tools/)
2
+
3
+ This package is intended to simplify the use of
4
+ [Model Context Protocol (MCP)](https://modelcontextprotocol.io/)
5
+ server tools with LangChain / Python.
6
+
7
+ It contains a utility function `convertMcpToLangchainTools()`.
8
+ This function handles parallel initialization of specified multiple MCP servers
9
+ and converts their available tools into an array of
10
+ [LangChain-compatible tools](https://js.langchain.com/docs/how_to/tool_calling/).
11
+
12
+ A typescript version of this utility library is available
13
+ [here](https://www.npmjs.com/package/@h1deya/langchain-mcp-tools)
14
+
15
+ ## Requirements
16
+
17
+ - Python 3.11+
18
+
19
+ ## Installation
20
+
21
+ ```bash
22
+ pip install langchain-mcp-tools
23
+ ```
24
+
25
+ ## Quick Start
26
+
27
+ `convert_mcp_to_langchain_tools()` utility function accepts MCP server configurations
28
+ that follow the same structure as
29
+ [Claude for Desktop](https://modelcontextprotocol.io/quickstart/user),
30
+ but only the contents of the `mcpServers` property,
31
+ and is expressed as a `dict`, e.g.:
32
+
33
+ ```python
34
+ mcp_configs = {
35
+ 'filesystem': {
36
+ 'command': 'npx',
37
+ 'args': ['-y', '@modelcontextprotocol/server-filesystem', '.']
38
+ },
39
+ 'fetch': {
40
+ 'command': 'uvx',
41
+ 'args': ['mcp-server-fetch']
42
+ }
43
+ }
44
+
45
+ tools, cleanup = await convert_mcp_to_langchain_tools(
46
+ mcp_configs
47
+ )
48
+ ```
49
+
50
+ The utility function initializes all specified MCP servers in parallel,
51
+ and returns LangChain Tools (`List[BaseTool]`)
52
+ by gathering all available MCP server tools,
53
+ and by wrapping them into [LangChain Tools](https://js.langchain.com/docs/how_to/tool_calling/).
54
+ It also returns a cleanup callback function (`McpServerCleanupFn`)
55
+ to be invoked to close all MCP server sessions when finished.
56
+
57
+ The returned tools can be used with LangChain, e.g.:
58
+
59
+ ```python
60
+ # from langchain.chat_models import init_chat_model
61
+ llm = init_chat_model(
62
+ model='claude-3-5-haiku-latest',
63
+ model_provider='anthropic'
64
+ )
65
+
66
+ # from langgraph.prebuilt import create_react_agent
67
+ agent = create_react_agent(
68
+ llm,
69
+ tools
70
+ )
71
+ ```
72
+ <!-- A simple and experimentable usage example can be found
73
+ [here](https://github.com/hideya/langchain-mcp-tools-ts-usage/blob/main/src/index.ts) -->
74
+
75
+ <!-- A more realistic usage example can be found
76
+ [here](https://github.com/hideya/langchain-mcp-client-ts) -->
77
+
78
+ An usage example can be found
79
+ [here](https://github.com/hideya/mcp-client-langchain-py)
80
+
81
+ ## Limitations
82
+
83
+ Currently, only text results of tool calls are supported.
84
+
85
+ ## Technical Details
86
+
87
+ It was very tricky (for me) to get the parallel MCP server initialization
88
+ to work successfully...
89
+
90
+ I'm new to Python, so it is very possible that my ignorance is playing
91
+ a big role here...
92
+ I'm summarizing why it was difficult for me below.
93
+ Any comments pointing out something I am missing would be greatly appreciated!
94
+ [(comment here)](https://github.com/hideya/langchain-mcp-tools-ts/issues)
95
+
96
+ 1. Core Challenge:
97
+ - Async resources management for `stdio_client` and `ClientSession` seems
98
+ to rely exclusively on `asynccontextmanager` for cleanup with no manual
99
+ cleanup options (based on the mcp python-sdk impl as of Jan 14, 2025)
100
+ - Initializing multiple MCP servers in parallel requires a dedicated
101
+ `asyncio.Task` per server
102
+ - Necessity of keeping sessions alive for later use after initialization
103
+ - Ensuring proper cleanup later in the same task that created them
104
+
105
+ 2. Solution Strategy:
106
+ A key requirement for parallel initialization is that each server must be
107
+ initialized in its own dedicated task - there's no way around this if we
108
+ want true parallel initialization. However, this creates a challenge since
109
+ we also need to maintain long-lived sessions and handle cleanup properly.
110
+
111
+ The key insight is to keep the initialization tasks alive throughout the
112
+ session lifetime, rather than letting them complete after initialization.
113
+
114
+ By using `asyncio.Event`s for coordination, we can:
115
+ - Allow parallel initialization while maintaining proper context management
116
+ - Keep each initialization task running until explicit cleanup is requested
117
+ - Ensure cleanup occurs in the same task that created the resources
118
+ - Provide a clean interface for the caller to manage the lifecycle
119
+
120
+ Alternative Considered:
121
+ A generator/coroutine approach using `finally` block for cleanup was
122
+ considered but rejected because:
123
+ - It turned out that the `finally` block in a generator/coroutine can be
124
+ executed by a different task than the one that ran the main body of
125
+ the code
126
+ - This breaks the requirement that `AsyncExitStack.aclose()` must be
127
+ called from the same task that created the context
128
+
129
+ 3. Task Lifecycle:
130
+ To allow the initialization task to stay alive waiting for cleanup:
131
+ ```
132
+ [Task starts]
133
+ ↓
134
+ Initialize server & convert tools
135
+ ↓
136
+ Set ready_event (signals tools are ready)
137
+ ↓
138
+ await cleanup_event.wait() (keeps task alive)
139
+ ↓
140
+ When cleanup_event is set:
141
+ exit_stack.aclose() (cleanup in original task)
142
+ ```
143
+ This approach indeed enables parallel initialization while maintaining proper
144
+ async resource lifecycle management through context managers.
145
+ However, I'm afraid I'm twisting things around too much.
146
+ It usually means I'm doing something very worng...
147
+
148
+ I think it is a natural assumption that MCP SDK is designed with consideration
149
+ for parallel server initialization.
150
+ I'm not sure what I'm missing...
151
+ (FYI, with the TypeScript SDK, parallel server initialization was pretty straightforward)
langchain_mcp_tools-0.0.4/langchain_mcp_tools.egg-info/PKG-INFO ADDED
@@ -0,0 +1,172 @@
1
+ Metadata-Version: 2.2
2
+ Name: langchain-mcp-tools
3
+ Version: 0.0.4
4
+ Summary: Model Context Protocol (MCP) To LangChain Tools Conversion Utility
5
+ Project-URL: Bug Tracker, https://github.com/hideya/langchain-mcp-tools-py/issues
6
+ Project-URL: Source Code, https://github.com/hideya/langchain-mcp-tools-py
7
+ Requires-Python: >=3.11
8
+ Description-Content-Type: text/markdown
9
+ License-File: LICENSE
10
+ Requires-Dist: jsonschema-pydantic>=0.6
11
+ Requires-Dist: langchain-core>=0.3.29
12
+ Requires-Dist: langchain>=0.3.14
13
+ Requires-Dist: langchain-anthropic>=0.3.1
14
+ Requires-Dist: langchain-groq>=0.2.3
15
+ Requires-Dist: langchain-openai>=0.3.0
16
+ Requires-Dist: langgraph>=0.2.62
17
+ Requires-Dist: mcp>=1.2.0
18
+ Requires-Dist: pyjson5>=1.6.8
19
+ Requires-Dist: pympler>=1.1
20
+ Requires-Dist: python-dotenv>=1.0.1
21
+
22
+ # MCP To LangChain Tools Conversion Utility / Python [![License: MIT](https://img.shields.io/badge/License-MIT-blue.svg)](https://github.com/hideya/mcp-langchain-client-ts/blob/main/LICENSE) [![pypi version](https://img.shields.io/pypi/v/langchain-mcp-tools.svg)](https://pypi.org/project/langchain-mcp-tools/)
23
+
24
+ This package is intended to simplify the use of
25
+ [Model Context Protocol (MCP)](https://modelcontextprotocol.io/)
26
+ server tools with LangChain / Python.
27
+
28
+ It contains a utility function `convertMcpToLangchainTools()`.
29
+ This function handles parallel initialization of specified multiple MCP servers
30
+ and converts their available tools into an array of
31
+ [LangChain-compatible tools](https://js.langchain.com/docs/how_to/tool_calling/).
32
+
33
+ A typescript version of this utility library is available
34
+ [here](https://www.npmjs.com/package/@h1deya/langchain-mcp-tools)
35
+
36
+ ## Requirements
37
+
38
+ - Python 3.11+
39
+
40
+ ## Installation
41
+
42
+ ```bash
43
+ pip install langchain-mcp-tools
44
+ ```
45
+
46
+ ## Quick Start
47
+
48
+ `convert_mcp_to_langchain_tools()` utility function accepts MCP server configurations
49
+ that follow the same structure as
50
+ [Claude for Desktop](https://modelcontextprotocol.io/quickstart/user),
51
+ but only the contents of the `mcpServers` property,
52
+ and is expressed as a `dict`, e.g.:
53
+
54
+ ```python
55
+ mcp_configs = {
56
+ 'filesystem': {
57
+ 'command': 'npx',
58
+ 'args': ['-y', '@modelcontextprotocol/server-filesystem', '.']
59
+ },
60
+ 'fetch': {
61
+ 'command': 'uvx',
62
+ 'args': ['mcp-server-fetch']
63
+ }
64
+ }
65
+
66
+ tools, cleanup = await convert_mcp_to_langchain_tools(
67
+ mcp_configs
68
+ )
69
+ ```
70
+
71
+ The utility function initializes all specified MCP servers in parallel,
72
+ and returns LangChain Tools (`List[BaseTool]`)
73
+ by gathering all available MCP server tools,
74
+ and by wrapping them into [LangChain Tools](https://js.langchain.com/docs/how_to/tool_calling/).
75
+ It also returns a cleanup callback function (`McpServerCleanupFn`)
76
+ to be invoked to close all MCP server sessions when finished.
77
+
78
+ The returned tools can be used with LangChain, e.g.:
79
+
80
+ ```python
81
+ # from langchain.chat_models import init_chat_model
82
+ llm = init_chat_model(
83
+ model='claude-3-5-haiku-latest',
84
+ model_provider='anthropic'
85
+ )
86
+
87
+ # from langgraph.prebuilt import create_react_agent
88
+ agent = create_react_agent(
89
+ llm,
90
+ tools
91
+ )
92
+ ```
93
+ <!-- A simple and experimentable usage example can be found
94
+ [here](https://github.com/hideya/langchain-mcp-tools-ts-usage/blob/main/src/index.ts) -->
95
+
96
+ <!-- A more realistic usage example can be found
97
+ [here](https://github.com/hideya/langchain-mcp-client-ts) -->
98
+
99
+ An usage example can be found
100
+ [here](https://github.com/hideya/mcp-client-langchain-py)
101
+
102
+ ## Limitations
103
+
104
+ Currently, only text results of tool calls are supported.
105
+
106
+ ## Technical Details
107
+
108
+ It was very tricky (for me) to get the parallel MCP server initialization
109
+ to work successfully...
110
+
111
+ I'm new to Python, so it is very possible that my ignorance is playing
112
+ a big role here...
113
+ I'm summarizing why it was difficult for me below.
114
+ Any comments pointing out something I am missing would be greatly appreciated!
115
+ [(comment here)](https://github.com/hideya/langchain-mcp-tools-ts/issues)
116
+
117
+ 1. Core Challenge:
118
+ - Async resources management for `stdio_client` and `ClientSession` seems
119
+ to rely exclusively on `asynccontextmanager` for cleanup with no manual
120
+ cleanup options (based on the mcp python-sdk impl as of Jan 14, 2025)
121
+ - Initializing multiple MCP servers in parallel requires a dedicated
122
+ `asyncio.Task` per server
123
+ - Necessity of keeping sessions alive for later use after initialization
124
+ - Ensuring proper cleanup later in the same task that created them
125
+
126
+ 2. Solution Strategy:
127
+ A key requirement for parallel initialization is that each server must be
128
+ initialized in its own dedicated task - there's no way around this if we
129
+ want true parallel initialization. However, this creates a challenge since
130
+ we also need to maintain long-lived sessions and handle cleanup properly.
131
+
132
+ The key insight is to keep the initialization tasks alive throughout the
133
+ session lifetime, rather than letting them complete after initialization.
134
+
135
+ By using `asyncio.Event`s for coordination, we can:
136
+ - Allow parallel initialization while maintaining proper context management
137
+ - Keep each initialization task running until explicit cleanup is requested
138
+ - Ensure cleanup occurs in the same task that created the resources
139
+ - Provide a clean interface for the caller to manage the lifecycle
140
+
141
+ Alternative Considered:
142
+ A generator/coroutine approach using `finally` block for cleanup was
143
+ considered but rejected because:
144
+ - It turned out that the `finally` block in a generator/coroutine can be
145
+ executed by a different task than the one that ran the main body of
146
+ the code
147
+ - This breaks the requirement that `AsyncExitStack.aclose()` must be
148
+ called from the same task that created the context
149
+
150
+ 3. Task Lifecycle:
151
+ To allow the initialization task to stay alive waiting for cleanup:
152
+ ```
153
+ [Task starts]
154
+ ↓
155
+ Initialize server & convert tools
156
+ ↓
157
+ Set ready_event (signals tools are ready)
158
+ ↓
159
+ await cleanup_event.wait() (keeps task alive)
160
+ ↓
161
+ When cleanup_event is set:
162
+ exit_stack.aclose() (cleanup in original task)
163
+ ```
164
+ This approach indeed enables parallel initialization while maintaining proper
165
+ async resource lifecycle management through context managers.
166
+ However, I'm afraid I'm twisting things around too much.
167
+ It usually means I'm doing something very worng...
168
+
169
+ I think it is a natural assumption that MCP SDK is designed with consideration
170
+ for parallel server initialization.
171
+ I'm not sure what I'm missing...
172
+ (FYI, with the TypeScript SDK, parallel server initialization was pretty straightforward)
{langchain_mcp_tools-0.0.2 → langchain_mcp_tools-0.0.4}/pyproject.toml RENAMED
@@ -1,7 +1,7 @@
1
1
  [project]
2
2
  name = "langchain-mcp-tools"
3
- version = "0.0.2"
4
- description = "MCP To LangChain Tools Conversion Utility"
3
+ version = "0.0.4"
4
+ description = "Model Context Protocol (MCP) To LangChain Tools Conversion Utility"
5
5
  readme = "README.md"
6
6
  requires-python = ">=3.11"
7
7
  dependencies = [
@@ -33,7 +33,5 @@ exclude = [
33
33
  ]
34
34
 
35
35
  [project.urls]
36
- "Homepage" = "https://github.com/hideya"
37
- "Bug Tracker" = "https://github.com/hideya/mcp-client-langchain-py/issues"
38
- "Documentation" = "https://github.com/hideya/mcp-client-langchain-py/blob/main/README.md"
39
- "Source Code" = "https://github.com/hideya/mcp-client-langchain-py"
36
+ "Bug Tracker" = "https://github.com/hideya/langchain-mcp-tools-py/issues"
37
+ "Source Code" = "https://github.com/hideya/langchain-mcp-tools-py"
{langchain_mcp_tools-0.0.2 → langchain_mcp_tools-0.0.4}/src/langchain_mcp_tools.py RENAMED
@@ -37,11 +37,12 @@ require context managers while enabling parallel initialization.
37
37
  The key aspects are:
38
38
 
39
39
  1. Core Challenge:
40
- - Managing async resources (stdio_client and ClientSession) that seems to
41
- rely exclusively on asynccontextmanager for cleanup with no manual cleanup
42
- options (based on the mcp python-sdk impl as of Jan 14, 2025 #62a0af6)
43
- - Initializing multiple servers in parallel
44
- - Keeping sessions alive for later use
40
+ - Async resources management for `stdio_client` and `ClientSession` seems
41
+ to rely exclusively on `asynccontextmanager` for cleanup with no manual
42
+ cleanup options (based on the mcp python-sdk impl as of Jan 14, 2025)
43
+ - Initializing multiple MCP servers in parallel requires a dedicated
44
+ `asyncio.Task` per server
45
+ - Necessity of keeping sessions alive for later use after initialization
45
46
  - Ensuring proper cleanup in the same task that created them
46
47
 
47
48
  2. Solution Strategy:
@@ -52,18 +53,19 @@ The key aspects are:
52
53
 
53
54
  The key insight is to keep the initialization tasks alive throughout the
54
55
  session lifetime, rather than letting them complete after initialization.
55
- By using events for coordination, we can:
56
+ By using `asyncio.Event`s for coordination, we can:
56
57
  - Allow parallel initialization while maintaining proper context management
57
58
  - Keep each initialization task running until explicit cleanup is requested
58
59
  - Ensure cleanup occurs in the same task that created the resources
59
60
  - Provide a clean interface for the caller to manage the lifecycle
60
61
 
61
62
  Alternative Considered:
62
- A generator/coroutine approach using 'finally' block for cleanup was
63
+ A generator/coroutine approach using `finally` block for cleanup was
63
64
  considered but rejected because:
64
- - The 'finally' block in a generator/coroutine can be executed by a
65
- different task than the one that ran the main body of the code
66
- - This breaks the requirement that AsyncExitStack.aclose() must be
65
+ - It turned out that the `finally` block in a generator/coroutine can be
66
+ executed by a different task than the one that ran the main body of
67
+ the code
68
+ - This breaks the requirement that `AsyncExitStack.aclose()` must be
67
69
  called from the same task that created the context
68
70
 
69
71
  3. Task Lifecycle:
@@ -79,8 +81,16 @@ The key aspects are:
79
81
  When cleanup_event is set:
80
82
  exit_stack.aclose() (cleanup in original task)
81
83
 
82
- This pattern enables parallel initialization while maintaining proper async
83
- resource lifecycle management through context managers.
84
+ This approach indeed enables parallel initialization while maintaining proper
85
+ async resource lifecycle management through context managers.
86
+ However, I'm afraid I'm twisting things around too much.
87
+ It usually means I'm doing something very worng...
88
+
89
+ I think it is a natural assumption that MCP SDK is designed with consideration
90
+ for parallel server initialization.
91
+ I'm not sure what I'm missing...
92
+ (FYI, with the TypeScript SDK, parallel server initializaion was quite
93
+ straight forward)
84
94
  """
85
95
 
86
96
 
langchain_mcp_tools-0.0.2/PKG-INFO DELETED
@@ -1,106 +0,0 @@
1
- Metadata-Version: 2.2
2
- Name: langchain-mcp-tools
3
- Version: 0.0.2
4
- Summary: MCP To LangChain Tools Conversion Utility
5
- Project-URL: Homepage, https://github.com/hideya
6
- Project-URL: Bug Tracker, https://github.com/hideya/mcp-client-langchain-py/issues
7
- Project-URL: Documentation, https://github.com/hideya/mcp-client-langchain-py/blob/main/README.md
8
- Project-URL: Source Code, https://github.com/hideya/mcp-client-langchain-py
9
- Requires-Python: >=3.11
10
- Description-Content-Type: text/markdown
11
- License-File: LICENSE
12
- Requires-Dist: jsonschema-pydantic>=0.6
13
- Requires-Dist: langchain-core>=0.3.29
14
- Requires-Dist: langchain>=0.3.14
15
- Requires-Dist: langchain-anthropic>=0.3.1
16
- Requires-Dist: langchain-groq>=0.2.3
17
- Requires-Dist: langchain-openai>=0.3.0
18
- Requires-Dist: langgraph>=0.2.62
19
- Requires-Dist: mcp>=1.2.0
20
- Requires-Dist: pyjson5>=1.6.8
21
- Requires-Dist: pympler>=1.1
22
- Requires-Dist: python-dotenv>=1.0.1
23
-
24
- # MCP To LangChain Tools Conversion Utility / Python [![License: MIT](https://img.shields.io/badge/License-MIT-blue.svg)](https://github.com/hideya/mcp-langchain-client-ts/blob/main/LICENSE) [![pypi version](https://img.shields.io/pypi/v/langchain-mcp-tools.svg)](https://pypi.org/project/langchain-mcp-tools/)
25
-
26
- This package is intended to simplify the use of
27
- [Model Context Protocol (MCP)](https://modelcontextprotocol.io/)
28
- server tools with LangChain / Python.
29
-
30
- It contains a utility function `convertMcpToLangchainTools()`.
31
- This function handles parallel initialization of specified multiple MCP servers
32
- and converts their available tools into an array of
33
- [LangChain-compatible tools](https://js.langchain.com/docs/how_to/tool_calling/).
34
-
35
- A typescript version of this MCP client is available
36
- [here](https://www.npmjs.com/package/@h1deya/langchain-mcp-tools)
37
-
38
- ## Requirements
39
-
40
- - Python 3.11+
41
-
42
- ## Installation
43
-
44
- ```bash
45
- pip install langchain-mcp-tools
46
- ```
47
-
48
- ## Quick Start
49
-
50
- `convert_mcp_to_langchain_tools()` utility function accepts MCP server configurations
51
- that follow the same structure as
52
- [Claude for Desktop](https://modelcontextprotocol.io/quickstart/user),
53
- but only the contents of the `mcpServers` property,
54
- and is expressed as a `dict`, e.g.:
55
-
56
- ```python
57
- mcp_configs = {
58
- 'filesystem': {
59
- 'command': 'npx',
60
- 'args': ['-y', '@modelcontextprotocol/server-filesystem', '.']
61
- },
62
- 'fetch': {
63
- 'command': 'uvx',
64
- 'args': ['mcp-server-fetch']
65
- }
66
- }
67
-
68
- tools, mcp_cleanup = await convert_mcp_to_langchain_tools(
69
- mcp_configs
70
- )
71
- ```
72
-
73
- The utility function initializes all specified MCP servers in parallel,
74
- and returns LangChain Tools (`List[BaseTool]`)
75
- by gathering all available MCP server tools,
76
- and by wrapping them into [LangChain Tools](https://js.langchain.com/docs/how_to/tool_calling/).
77
- It also returns a cleanup callback function (`McpServerCleanupFn`)
78
- to be invoked to close all MCP server sessions when finished.
79
-
80
- The returned tools can be used with LangChain, e.g.:
81
-
82
- ```python
83
- # from langchain.chat_models import init_chat_model
84
- llm = init_chat_model(
85
- model='claude-3-5-haiku-latest',
86
- model_provider='anthropic'
87
- )
88
-
89
- # from langgraph.prebuilt import create_react_agent
90
- agent = create_react_agent(
91
- llm,
92
- tools
93
- )
94
- ```
95
- <!-- A simple and experimentable usage example can be found
96
- [here](https://github.com/hideya/langchain-mcp-tools-ts-usage/blob/main/src/index.ts) -->
97
-
98
- <!-- A more realistic usage example can be found
99
- [here](https://github.com/hideya/langchain-mcp-client-ts) -->
100
-
101
- An usage example can be found
102
- [here](https://github.com/hideya/mcp-client-langchain-py)
103
-
104
- ## Limitations
105
-
106
- Currently, only text results of tool calls are supported.
langchain_mcp_tools-0.0.2/README.md DELETED
@@ -1,83 +0,0 @@
1
- # MCP To LangChain Tools Conversion Utility / Python [![License: MIT](https://img.shields.io/badge/License-MIT-blue.svg)](https://github.com/hideya/mcp-langchain-client-ts/blob/main/LICENSE) [![pypi version](https://img.shields.io/pypi/v/langchain-mcp-tools.svg)](https://pypi.org/project/langchain-mcp-tools/)
2
-
3
- This package is intended to simplify the use of
4
- [Model Context Protocol (MCP)](https://modelcontextprotocol.io/)
5
- server tools with LangChain / Python.
6
-
7
- It contains a utility function `convertMcpToLangchainTools()`.
8
- This function handles parallel initialization of specified multiple MCP servers
9
- and converts their available tools into an array of
10
- [LangChain-compatible tools](https://js.langchain.com/docs/how_to/tool_calling/).
11
-
12
- A typescript version of this MCP client is available
13
- [here](https://www.npmjs.com/package/@h1deya/langchain-mcp-tools)
14
-
15
- ## Requirements
16
-
17
- - Python 3.11+
18
-
19
- ## Installation
20
-
21
- ```bash
22
- pip install langchain-mcp-tools
23
- ```
24
-
25
- ## Quick Start
26
-
27
- `convert_mcp_to_langchain_tools()` utility function accepts MCP server configurations
28
- that follow the same structure as
29
- [Claude for Desktop](https://modelcontextprotocol.io/quickstart/user),
30
- but only the contents of the `mcpServers` property,
31
- and is expressed as a `dict`, e.g.:
32
-
33
- ```python
34
- mcp_configs = {
35
- 'filesystem': {
36
- 'command': 'npx',
37
- 'args': ['-y', '@modelcontextprotocol/server-filesystem', '.']
38
- },
39
- 'fetch': {
40
- 'command': 'uvx',
41
- 'args': ['mcp-server-fetch']
42
- }
43
- }
44
-
45
- tools, mcp_cleanup = await convert_mcp_to_langchain_tools(
46
- mcp_configs
47
- )
48
- ```
49
-
50
- The utility function initializes all specified MCP servers in parallel,
51
- and returns LangChain Tools (`List[BaseTool]`)
52
- by gathering all available MCP server tools,
53
- and by wrapping them into [LangChain Tools](https://js.langchain.com/docs/how_to/tool_calling/).
54
- It also returns a cleanup callback function (`McpServerCleanupFn`)
55
- to be invoked to close all MCP server sessions when finished.
56
-
57
- The returned tools can be used with LangChain, e.g.:
58
-
59
- ```python
60
- # from langchain.chat_models import init_chat_model
61
- llm = init_chat_model(
62
- model='claude-3-5-haiku-latest',
63
- model_provider='anthropic'
64
- )
65
-
66
- # from langgraph.prebuilt import create_react_agent
67
- agent = create_react_agent(
68
- llm,
69
- tools
70
- )
71
- ```
72
- <!-- A simple and experimentable usage example can be found
73
- [here](https://github.com/hideya/langchain-mcp-tools-ts-usage/blob/main/src/index.ts) -->
74
-
75
- <!-- A more realistic usage example can be found
76
- [here](https://github.com/hideya/langchain-mcp-client-ts) -->
77
-
78
- An usage example can be found
79
- [here](https://github.com/hideya/mcp-client-langchain-py)
80
-
81
- ## Limitations
82
-
83
- Currently, only text results of tool calls are supported.
langchain_mcp_tools-0.0.2/langchain_mcp_tools.egg-info/PKG-INFO DELETED
@@ -1,106 +0,0 @@
1
- Metadata-Version: 2.2
2
- Name: langchain-mcp-tools
3
- Version: 0.0.2
4
- Summary: MCP To LangChain Tools Conversion Utility
5
- Project-URL: Homepage, https://github.com/hideya
6
- Project-URL: Bug Tracker, https://github.com/hideya/mcp-client-langchain-py/issues
7
- Project-URL: Documentation, https://github.com/hideya/mcp-client-langchain-py/blob/main/README.md
8
- Project-URL: Source Code, https://github.com/hideya/mcp-client-langchain-py
9
- Requires-Python: >=3.11
10
- Description-Content-Type: text/markdown
11
- License-File: LICENSE
12
- Requires-Dist: jsonschema-pydantic>=0.6
13
- Requires-Dist: langchain-core>=0.3.29
14
- Requires-Dist: langchain>=0.3.14
15
- Requires-Dist: langchain-anthropic>=0.3.1
16
- Requires-Dist: langchain-groq>=0.2.3
17
- Requires-Dist: langchain-openai>=0.3.0
18
- Requires-Dist: langgraph>=0.2.62
19
- Requires-Dist: mcp>=1.2.0
20
- Requires-Dist: pyjson5>=1.6.8
21
- Requires-Dist: pympler>=1.1
22
- Requires-Dist: python-dotenv>=1.0.1
23
-
24
- # MCP To LangChain Tools Conversion Utility / Python [![License: MIT](https://img.shields.io/badge/License-MIT-blue.svg)](https://github.com/hideya/mcp-langchain-client-ts/blob/main/LICENSE) [![pypi version](https://img.shields.io/pypi/v/langchain-mcp-tools.svg)](https://pypi.org/project/langchain-mcp-tools/)
25
-
26
- This package is intended to simplify the use of
27
- [Model Context Protocol (MCP)](https://modelcontextprotocol.io/)
28
- server tools with LangChain / Python.
29
-
30
- It contains a utility function `convertMcpToLangchainTools()`.
31
- This function handles parallel initialization of specified multiple MCP servers
32
- and converts their available tools into an array of
33
- [LangChain-compatible tools](https://js.langchain.com/docs/how_to/tool_calling/).
34
-
35
- A typescript version of this MCP client is available
36
- [here](https://www.npmjs.com/package/@h1deya/langchain-mcp-tools)
37
-
38
- ## Requirements
39
-
40
- - Python 3.11+
41
-
42
- ## Installation
43
-
44
- ```bash
45
- pip install langchain-mcp-tools
46
- ```
47
-
48
- ## Quick Start
49
-
50
- `convert_mcp_to_langchain_tools()` utility function accepts MCP server configurations
51
- that follow the same structure as
52
- [Claude for Desktop](https://modelcontextprotocol.io/quickstart/user),
53
- but only the contents of the `mcpServers` property,
54
- and is expressed as a `dict`, e.g.:
55
-
56
- ```python
57
- mcp_configs = {
58
- 'filesystem': {
59
- 'command': 'npx',
60
- 'args': ['-y', '@modelcontextprotocol/server-filesystem', '.']
61
- },
62
- 'fetch': {
63
- 'command': 'uvx',
64
- 'args': ['mcp-server-fetch']
65
- }
66
- }
67
-
68
- tools, mcp_cleanup = await convert_mcp_to_langchain_tools(
69
- mcp_configs
70
- )
71
- ```
72
-
73
- The utility function initializes all specified MCP servers in parallel,
74
- and returns LangChain Tools (`List[BaseTool]`)
75
- by gathering all available MCP server tools,
76
- and by wrapping them into [LangChain Tools](https://js.langchain.com/docs/how_to/tool_calling/).
77
- It also returns a cleanup callback function (`McpServerCleanupFn`)
78
- to be invoked to close all MCP server sessions when finished.
79
-
80
- The returned tools can be used with LangChain, e.g.:
81
-
82
- ```python
83
- # from langchain.chat_models import init_chat_model
84
- llm = init_chat_model(
85
- model='claude-3-5-haiku-latest',
86
- model_provider='anthropic'
87
- )
88
-
89
- # from langgraph.prebuilt import create_react_agent
90
- agent = create_react_agent(
91
- llm,
92
- tools
93
- )
94
- ```
95
- <!-- A simple and experimentable usage example can be found
96
- [here](https://github.com/hideya/langchain-mcp-tools-ts-usage/blob/main/src/index.ts) -->
97
-
98
- <!-- A more realistic usage example can be found
99
- [here](https://github.com/hideya/langchain-mcp-client-ts) -->
100
-
101
- An usage example can be found
102
- [here](https://github.com/hideya/mcp-client-langchain-py)
103
-
104
- ## Limitations
105
-
106
- Currently, only text results of tool calls are supported.