cloudparts

Glama

The largest public directory of MCP servers, which builds and runs each open-source server in a Firecracker sandbox to record its tools, score their descriptions and flag risky behavior, plus hosting and a gateway for the servers you use.

What it is
Agent tools and integrations
Type
Commercial product
Made by
Glama LLC
License
Proprietary
Facts checked
13 October 2026
Website

What it is

Glama lists about 97,600 open-source MCP servers and 30,400 remote connectors, with some 985,000 tools between them, and re-publishes everything in the official MCP Registry. A maintainer claims a listing by signing in with GitHub and proving write access to the repository. Glama then clones the repository, builds it from its Dockerfile (or one its build system infers), and runs it in a throwaway Firecracker microVM. Inside, it asks the server for its tools, resources and prompts and stores their full JSON schemas, while watching system calls and network traffic for things like reads of SSH keys or traffic to hosts the code never mentions. Servers that fail to build stay hidden from search, and every new commit triggers a rescan; Glama says it ran over a million scans in a year.

Each tool description gets a Tool Definition Quality Score across six dimensions (purpose, usage guidelines, side effects, parameters, conciseness and completeness), and remote connectors are re-introspected on a schedule, with schema changes diffed and checked for prompt injection. The results are public on each listing and through a REST API.

Around the directory Glama sells the pieces for running servers: one-click hosting on dedicated machines, a gateway that sits between clients and any MCP server to manage OAuth tokens, allow or block single tools and log every JSON-RPC message, and a browser-based MCP inspector. Glama also runs a chat workspace and an AI gateway for model providers; in April 2026 it open-sourced the gateway’s core as Lightport, limited the hosted gateway to a curated set of providers and said it was going all-in on MCP.

Why teams pick it

  • Breadth: about 97,600 servers and 30,400 connectors, a superset of the official MCP Registry
  • Evidence instead of a bare listing: captured tool schemas, sandbox behavior findings, quality scores and change history for each server
  • A machine-readable API (with an OpenAPI description) and per-server config snippets for Cursor, Claude Desktop, Claude Code and VS Code
  • One vendor for finding, testing, hosting and gating servers, with call logs and per-tool access control in the gateway

Watch out for

  • Directory API reads need an API key, and the data license requires visible attribution to Glama and a link to each record’s listing wherever you show it
  • Scans are automated heuristics: Glama keeps its detection rules partly private, and a clean listing is not a security review of your own deployment
  • Prices for hosting, the gateway and chat are behind sign-up rather than on a public pricing page
  • The pipeline is built and maintained by Frank Fiegel and contributors, and the directory competes with copycat sites and with the free official MCP Registry

Alternatives

What all agent tools and integrations have in common →Compare side by side →
ComposioCommercialA hosted tool layer for AI agents: one session per end user gives the model a few meta tools to search, authorize and run actions across more than 1,500 app toolkits, with Composio holding and refreshing the OAuth tokens.Model Context Protocol (MCP)Open sourceAn open protocol, built on JSON-RPC, that lets an AI application discover and call tools, read resources and use prompt templates from any server that speaks it, so one integration works across Claude, ChatGPT, Cursor, VS Code and other clients.OneRenamedCommercialA hosted connection layer for AI agents, formerly Pica: you connect each app once in your own account, and every agent reaches it through one MCP server or CLI, with per-action scopes enforced and logged on each call.OpenAPI SearchOpen sourceA small, open directory of OpenAPI descriptions for popular APIs, with an MCP server that turns any of them, or any spec URL, into a short plain-language summary an AI model can read before writing calls.SpeakeasyCommercialGenerates typed SDKs, Terraform providers, CLIs and MCP servers from an OpenAPI document, and now also sells a hosted MCP platform and an enterprise MCP gateway that governs which agents may call which tools.

How it fits with other technologies

Search the directory through the MCP API · bash

# Directory reads need an API key from glama.ai/settings/api-keys
curl -s "https://glama.ai/api/mcp/v1/servers?query=postgres&first=5&sort=search-relevance:desc" \
  -H "Authorization: Bearer $GLAMA_API_KEY"

# Remote servers reachable over Streamable HTTP
curl -s "https://glama.ai/api/mcp/v1/connectors?query=linear" \
  -H "Authorization: Bearer $GLAMA_API_KEY"

# A server hosted on Glama, or routed through its gateway, gets one endpoint for any client:
#   https://glama.ai/endpoints/<your-connection-profile>/mcp

See it in a real stack

No architecture uses Glama yet. Drop it into the playground and wire it up yourself.Open playground →
Where it came from2024–2026

Anthropic published the Model Context Protocol in November 2024, and within days Glama’s blog ran an MCP quickstart by the author punkpeye (25 November), who also started the awesome-mcp-servers list on GitHub (30 November). The list, which links to Glama’s directory and its Discord, has since passed 96,000 stars on GitHub.

A list of links could not say whether a server worked or was safe, so Glama turned the directory into a registry: build every server from source in a sandbox, record what it actually exposes, and watch what it does. By late 2025 punkpeye wrote of a year of constant iteration on the directory, and of finding it copied page for page by a competitor. By 2026 Glama re-published everything in the official MCP Registry alongside its own findings, and in April it published its Tool Definition Quality Score as open source.

Glama had also run a chat workspace and an AI gateway in front of dozens of model providers. In April 2026 it open-sourced that gateway as Lightport, cut the hosted gateway down to a curated set of providers and pointed users to OpenRouter for the rest, saying it was going “all-in on the MCP ecosystem”.

Why it’s built this way

Build and run every server instead of trusting its README
Introspecting a running server gives the real tool schemas, and running it in a fresh Firecracker microVM lets Glama watch for credential access or unexpected network traffic without exposing shared infrastructure.
Score tool descriptions, not just servers
Models choose tools from their descriptions; research Glama cites found well-described tools were selected far more often and that most public tools had at least one defect, so each listing shows a per-tool score with fixes.
Be a superset of the official registry
The MCP steering group’s registry holds vendor-neutral metadata only, so Glama ingests all of it and layers its sandbox findings, scores and change history on top.
A gateway rather than a thin proxy
Teams running many servers need OAuth handled, tools enabled one by one and every call logged, which a pass-through proxy does not do.

Timeline

  1. 2024

    MCP quickstart on Glama’s blog on 25 November, days after MCP’s release

  2. 2024

    awesome-mcp-servers list started on GitHub on 30 November

  3. 2025

    punkpeye writes of a year of constant iteration on the directory in November

  4. 2026

    Tool Definition Quality Score published in April

  5. 2026

    AI gateway open-sourced as Lightport on 27 April; focus moves to MCP

Sources8 · facts checked 13 October 2026

Improve this page on GitHub