Managing MCP Tool Versions Like a Pro
MCP tools evolve. New features, bug fixes, breaking changes. This tutorial covers version management on mcpm.
Understanding MCP Versioning
mcpm follows semantic versioning (semver):
1.2.3
│ │ │
│ │ └─ Patch: bug fixes, no behavior change
│ └─── Minor: new features, backward compatible
└───── Major: breaking changes
Checking Your Installed Versions
mcpm-dev list
Output:
weather-mcp v2.1.0
database-explorer v1.5.2
github-assistant v0.9.1
slack-bot v3.0.0
Checking for Updates
mcpm-dev outdated
Output:
Package Current Latest Type
weather-mcp 2.1.0 2.1.1 patch
database-explorer 1.5.2 2.0.0 major ⚠️
github-assistant 0.9.1 0.9.1 up to date
slack-bot 3.0.0 3.1.0 minor
Updating Tools
Update Everything (Safe)
mcpm-dev update
This updates all tools to their latest compatible version (patch and minor bumps only — no breaking changes).
Update a Specific Tool
mcpm-dev update weather-mcp
Force Update (Including Major)
mcpm-dev update database-explorer --force
Use with caution — major versions may break your agent workflows.
Pinning Versions
Need stability? Pin a specific version:
mcpm-dev add weather-mcp@2.1.0
Or in your project's .mcpmrc:
{
"packages": {
"weather-mcp": "2.1.0",
"database-explorer": "~1.5.0",
"github-assistant": "^0.9.0"
}
}
Version ranges:
2.1.0 — exact version only
~1.5.0 — patch updates (1.5.x)
^0.9.0 — minor updates (0.x.x)
* — always latest (not recommended)
Handling Breaking Changes
When a tool has a major update:
1. Read the Changelog
mcpm-dev info database-explorer --changelog
2. Test in Isolation
Update in a test environment first:
npx -y database-explorer@2.0.0
3. Check Your Agent Configs
If the tool's API changed, update your agent prompts and configurations.
4. Update Gradually
mcpm-dev update database-explorer --force
Rolling Back
mcpm-dev update weather-mcp@2.1.0
This downgrades (or upgrades) to the specified version.
Automating Updates
Add to CI/CD:
- name: Check MCP updates
run: |
mcpm-dev outdated --json > outdated.json
# Alert if major updates available
Best Practices
- Pin in production — exact versions for reliability
- Test updates — don't blindly update all
- Read changelogs — before every major bump
- Automate patch updates — they're safe
- Commit
.mcpmrc — version control your tool dependencies
Version management isn't glamorous, but it keeps your AI agents reliable.
#MCP #Tutorial #Versioning #mcpm #DevOps
Managing MCP Tool Versions Like a Pro
MCP tools evolve. New features, bug fixes, breaking changes. This tutorial covers version management on mcpm.
Understanding MCP Versioning
mcpm follows semantic versioning (semver):
Checking Your Installed Versions
Output:
Checking for Updates
Output:
Updating Tools
Update Everything (Safe)
This updates all tools to their latest compatible version (patch and minor bumps only — no breaking changes).
Update a Specific Tool
Force Update (Including Major)
Use with caution — major versions may break your agent workflows.
Pinning Versions
Need stability? Pin a specific version:
Or in your project's
.mcpmrc:{ "packages": { "weather-mcp": "2.1.0", "database-explorer": "~1.5.0", "github-assistant": "^0.9.0" } }Version ranges:
2.1.0— exact version only~1.5.0— patch updates (1.5.x)^0.9.0— minor updates (0.x.x)*— always latest (not recommended)Handling Breaking Changes
When a tool has a major update:
1. Read the Changelog
2. Test in Isolation
Update in a test environment first:
# Test the new version npx -y database-explorer@2.0.03. Check Your Agent Configs
If the tool's API changed, update your agent prompts and configurations.
4. Update Gradually
mcpm-dev update database-explorer --force # Test thoroughly # Then update the restRolling Back
This downgrades (or upgrades) to the specified version.
Automating Updates
Add to CI/CD:
# .github/workflows/mcp-updates.yml - name: Check MCP updates run: | mcpm-dev outdated --json > outdated.json # Alert if major updates availableBest Practices
.mcpmrc— version control your tool dependenciesVersion management isn't glamorous, but it keeps your AI agents reliable.
#MCP #Tutorial #Versioning #mcpm #DevOps