mcp-proxy 0.3.1

Standalone MCP proxy -- config-driven reverse proxy with auth, rate limiting, and observability
Documentation 
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115

# Cost Optimization Proxy
#
# Controls AI tool usage costs by rate limiting expensive backends,
# caching aggressively, and limiting concurrency. Useful when backends
# call paid APIs (LLMs, search engines, cloud services).
#
# See: docs/architectures.md #11

[proxy]
name = "cost-optimizer"
version = "1.0.0"
separator = "/"
instructions = "Cost-optimized proxy. Expensive tool calls are rate-limited and cached."

[proxy.listen]
host = "0.0.0.0"
port = 8080

# --- Backends ---

# Free/cheap backend -- minimal restrictions
[[backends]]
name = "local-docs"
transport = "stdio"
command = "/usr/local/bin/docs-mcp-server"
args = ["--index", "/var/data/docs"]

[backends.cache]
resource_ttl_seconds = 3600  # 1 hour
tool_ttl_seconds = 600       # 10 minutes
max_entries = 10000

# Moderate cost -- search API with per-query pricing
[[backends]]
name = "search"
transport = "http"
url = "http://search-mcp.internal:8080"

[backends.rate_limit]
requests = 10
period_seconds = 60  # 10 req/min

[backends.cache]
resource_ttl_seconds = 300
tool_ttl_seconds = 120
max_entries = 5000

[backends.timeout]
seconds = 15

[backends.circuit_breaker]
failure_rate_threshold = 0.5
minimum_calls = 5
wait_duration_seconds = 120

# High cost -- LLM-backed summarization/analysis
[[backends]]
name = "ai-analysis"
transport = "http"
url = "http://analysis-mcp.internal:8080"
expose_tools = ["summarize", "classify", "extract_entities"]

[backends.rate_limit]
requests = 5
period_seconds = 60  # 5 req/min -- expensive calls

[backends.concurrency]
max_concurrent = 2

[backends.cache]
resource_ttl_seconds = 0
tool_ttl_seconds = 300  # Cache analysis results 5 min
max_entries = 1000

[backends.timeout]
seconds = 120  # LLM calls can be slow

[backends.retry]
max_retries = 1
initial_backoff_ms = 1000
max_backoff_ms = 5000

# Very high cost -- cloud infrastructure operations
[[backends]]
name = "cloud"
transport = "http"
url = "http://cloud-mcp.internal:8080"
expose_tools = ["list_instances", "get_costs", "get_metrics"]
hide_tools = ["create_instance", "delete_instance", "resize_instance"]

[backends.rate_limit]
requests = 2
period_seconds = 60  # 2 req/min

[backends.concurrency]
max_concurrent = 1

[backends.timeout]
seconds = 30

# --- Observability ---

[performance]
coalesce_requests = true

[security]
max_argument_size = 262144  # 256KB

[observability]
audit = true
log_level = "info"
json_logs = true

[observability.metrics]
enabled = true