glif-mcp-host
Operate sophisticated artificial intelligence procedures and maintain user profiles seamlessly, leveraging extensive glif resource specifics and associated metadata via the Model Context Protocol. Facilitate configuration of AI task tooling by dynamically integrating or retiring components as required.
Author

glifxyz
Quick Info
Actions
Tags
glif-mcp-host
Model Context Protocol (MCP) gateway for executing glif agent procedures originating from glif.app.
This backend service furnishes capabilities for initiating glif computations, administering autonomous agents (bots), and retrieving glif system metadata through the standardized Model Context Protocol (MCP).
Furthermore, this service supports the dynamic modification of the accessible toolset via meta-utilities such as add-tool and remove-tool. This includes the potentially radical capability to provision entire functional glif agents as a composite set of tools (complete with assigned persona). Note: This feature is currently in a highly experimental phase.
For comprehensive details, explore https://glif.app or engage with our community on Discord: https://discord.gg/glif
Core Capabilities
- Initiate glif executions with specified input parameters
- Retrieve in-depth diagnostics concerning glifs, execution instances, and associated user entities
- Access glif configuration metadata via resolvable URI paths
Deployment Instructions
Execution via npx (Preferred Method)
Assuming Node.js is available on your system, the package @glifxyz/glif-mcp-server can be invoked using npx:
- Secure your unique API credential from https://glif.app/settings/api-tokens
- Integrate the server configuration into your Claude Desktop configuration file. On macOS systems, this file resides at:
~/Library/Application Support/Claude/claude_desktop_config.json
{ "mcpServers": { "glif": { "command": "npx", "args": ["-y", "@glifxyz/glif-mcp-server@latest"], "env": { "GLIF_API_TOKEN": "your-token-here" } } } }
Execution from Source Repository Clone
Begin by cloning the repository and installing necessary prerequisites.
sh git clone https://github.com/glifxyz/glif-mcp-server cd glif-mcp-server npm install npm run build
The executable artifact is now located at build/index.js
Subsequently, configure your MCP client (e.g., Claude Desktop) to reference this locally compiled server binary.
{ "mcpServers": { "glif": { "command": "node", "args": ["/path/to/glif-mcp/build/index.js"], "env": { "GLIF_API_TOKEN": "your-token-here" } } } }
Optionally, you can supply a comma-delimited sequence of glif identifiers that should be pre-loaded upon server initialization. This is advantageous for controlled testing scenarios or distributing pre-configured glif setups.
{ "mcpServers": { "glif": { "command": "node", "args": ["/path/to/glif-mcp/build/index.js"], "env": { "GLIF_API_TOKEN": "your-token-here", "GLIF_IDS": "cm2v9aiga00008vfqdiximl2m,cm2v98jk6000r11afslqvooil,cm2v9rp66000bat9wr606qq6o", "IGNORE_SAVED_GLIFS": true, } } } }
Remote Operation via Smithery
To facilitate automated deployment and hosting of glif-mcp for Claude Desktop using Smithery, which manages the server infrastructure:
bash npx -y @smithery/cli install @glifxyz/glif-mcp-server --client claude
Operational Constraints
- Adherence to standard user account rate limitations applies.
- Capacity augmentation is available via https://glif.app/pricing
Resource Locators
glif://{id}- Access metadata pertaining to a specific glifglifRun://{id}- Retrieve execution metrics for a given run instanceglifUser://{id}- Fetch configuration and details for a user profile
Configuration Directives
Environment variables to govern the activation status of specific tool groupings:
GLIF_API_TOKEN- Mandatory. Your designated credential obtained from https://glif.app/settings/api-tokensGLIF_IDS- Optional. A delimited string of glif identifiers to be provisioned as initial tools.IGNORE_DISCOVERY_TOOLS- Set totrueto suppress discovery utilities (enabled by default)IGNORE_METASKILL_TOOLS- Set totrueto suppress meta-skill manipulation utilities (enabled by default)IGNORE_SAVED_GLIFS- Set totrueto prevent loading of persistently saved glif tools (enabled by default)BOT_TOOLS- Set totrueto activate autonomous agent tooling capabilities (disabled by default)
Toolset Inventory
Fundamental Utilities (Always Active)
run_glif- Execute a designated glif using provided input parametersglif_info- Retrieve comprehensive specifications for a glif, including required input schema
Discovery Utilities (Active by Default; Deactivate with IGNORE_DISCOVERY_TOOLS=true)
list_featured_glifs- Fetch a curated catalog of highlighted glifssearch_glifs- Perform textual searches across glif names and descriptionsmy_glifs- Obtain a roster of glifs owned by the authenticated principal (excluding draft versions)my_glif_user_info- Retrieve detailed data regarding the user account, recent executions, and frequently utilized glifs
Meta-Skill Utilities (Active by Default; Deactivate with IGNORE_METASKILL_TOOLS=true)
save_glif_as_tool- Persist a specific glif configuration as a new, dedicated toolremove_glif_tool- Decommission a previously saved glif-derived toolremove_all_glif_tools- Purge all personalized glif tools, restoring the default statelist_saved_glif_tools- Enumerate all currently registered custom glif tools
Autonomous Agent Utilities (Inactive by Default; Activate with BOT_TOOLS=true)
list_bots- Retrieve a directory of featured agents and simulation blueprintsload_bot- Fetch detailed specifications for a particular agent, including its repertoire of skillssave_bot_skills_as_tools- Convert all operational skills of a bot into distinct, callable toolsshow_bot_info- Display comprehensive details for a specified autonomous entity
Saved Glif Implementations (Active by Default; Deactivate with IGNORE_SAVED_GLIFS=true)
Dynamically generated utilities derived from glifs saved via the meta-skill interface. Every saved glif manifests as an independent tool, possessing a custom invocation name and descriptive text.
Process for Converting Glifs to Custom Tooling
While the generic run_glif utility exists, it suffers from two drawbacks: (a) its description is generic, and (b) it necessitates a preceding glif_info inquiry to ascertain the correct invocation signature. Moreover, manual knowledge of the glif's existence is required.
We are actively developing a suite of novel meta-tools designed to transform specific glifs into self-contained, standalone tools:
Illustrative interaction sequence:
- "What innovative glifs have been introduced recently?"
- [toolcall:
list_featured_glifs...] - "I approve of the 1970s science fiction novel cover generator; transform it into a specialized utility named \"scifi_book_image\""
- [toolcall:
save_glif_as_tool glifId=... toolName=scifi_book_image] - [The user can subsequently invoke the action simply by typing "render a sci-fi book image depicting X, Y, and Z"]
You can review these bespoke utilities using list_saved_glif_tools and discard unwanted ones via remove_glif_tool.
Note: Clients like Claude Desktop typically necessitate a process restart for the newly defined tool interfaces to be recognized. Clients such as Cline & Cursor often achieve automatic detection and re-querying of available toolsets upon modification.
Information access regarding the authenticated user's glif inventory:
my_glifs- Published glifs owned by the user (excludes drafts)my_liked_glifs- Glifs marked as favorites by the usermy_runs- Publicly accessible execution logs associated with the user
MCP Registry Links
Development Lifecycle
Prerequisite installation:
bash npm install
Building the server binary:
bash npm run build
For development workflows incorporating automatic rebuild triggers:
bash npm run dev
To execute the complete verification suite:
bash npm run test
And to maintain continuous testing upon file modification:
bash npm run test:watch
Troubleshooting Execution
Given that MCP servers communicate via standard input/output streams, debugging presents inherent difficulties. We strongly advise utilizing the MCP Inspector:
bash npm run inspector
The Inspector utility will furnish a network address to access diagnostic instrumentation within a web browser environment.
If utilizing Claude Desktop, diagnostic output streams can be directly inspected within the application's log files.
Issuing New Releases
- Modify version metadata in
package.jsonandsrc/index.tsto reflect the incremented release number. - Execute
npm installto synchronize version information in the lockfile. - Commit and push changes to the primary repository branch, merging into
main. - If the
ghCLI tool is installed, navigate tomainand executenpm run release. This command automates the creation of a git tag, pushes it to GitHub, and usesgh release createto publish the new version accompanied by an auto-generated release summary. Ifghis unavailable, these final steps must be completed manually via the GitHub web interface. - A designated GitHub Action will subsequently leverage the
NPM_TOKENsecret to deploy the package to the NPM registry.
Licensing
This software is distributed under the terms of the MIT License. Full details are available in the LICENSE file.
