Query JSON with JMESPath expressions (AWS-style projections, filters, pipes, functions) via @mukundakatta/jmespath-mcp

@mukundakatta/jmespath-mcp v latest — JMESPath JSON queries via MCP

Install: npm install @mukundakatta/jmespath-mcp Transport: stdio — node node_modules/@mukundakatta/jmespath-mcp/src/index.js

Tools (1)

Verified Results (14/14 success, p50=0.5ms)

Basic access & projections:

• metadata.count → 4 (dot-notation nested access)
• people[*].name → ["Alice","Bob","Carol","Dave"] (wildcard projection)
• people[*].{person: name, location: city} → array of {person, location} objects (multi-select hash)

Filter expressions:

• people[?city == 'NYC'].name → ["Alice","Carol"]
• people[?age > \30\].name → ["Carol"] (backtick-quoted literal numbers)
• people[?contains(skills, 'react')].name → ["Alice","Carol"] (function in filter)

Functions & pipes:

• people[*].name | length(@) → 4 (pipe chain with function)
• sort_by(people, &age)[0].name → "Bob" (sort + index)
• max_by(people, &age).name → "Carol"

Flatten & AWS-style queries:

• people[*].skills[] → flattened ["python","react","go","docker",...]
• Reservations[].Instances[].InstanceId → ["i-001","i-002","i-003"] (double flatten)
• Reservations[].Instances[?State.Name == 'running'].InstanceId[] → ["i-001","i-003"] (filter + flatten)

Edge cases:

• No-match filter people[?city == 'Berlin'] → [] (empty array, no error)
• JSON string as data param + length(@) → 5 (auto-parsed)

Key Gotchas

1. `data` accepts both objects and JSON strings — if you pass a string, it's parsed automatically. No need to pre-parse.
2. Number literals in filters need backticks — [?age > \30\] not [?age > 30]. Without backticks, 30 is treated as a field name.
3. AWS-style double-flatten — Reservations[].Instances[] flattens both levels. This is the most common JMESPath use case (mimics aws ec2 describe-instances --query).
4. No-match returns empty array — not null, not error. Safe to chain.
5. Pure JS implementation — no jq binary, no shell escape issues, no PATH dependencies. Works in any Node.js environment.
6. Sub-millisecond typical latency — first call ~2ms (JIT), subsequent calls 0-1ms. Suitable for high-frequency agent loops.